This talk addresses the overwhelming flood of information in the AI era by proposing a personal architecture of attention built through a deliberate writing practice. Rather than relying on rigid productivity systems, I share how combining morning voice dictation with an iterative AI workflow helps metabolize information and develop clear thinking. Ultimately, it encourages listeners to design their own structures for judgment so that they can navigate technological acceleration (or just life in general) with agency.

Recording:

Slides and Transcript:

Hello and Thank you for joining. By the end of this talk I want you to leave with a buildable practice — a personal architecture of agency: a deliberate structure for attention and judgment that AI can support but never replace.

Now I’ll start with a confession, because the last thing anyone needs is more AI hype.

I lead design for an AI agent system at Salesforce — which means I spend my days inside the acceleration most people are watching from outside. New interaction patterns, new tools, new capabilities.

The confession part: I cannot really keep up. Not with the reading, the model releases, or the discourse. I don’t think most people can — including the people whose job it is to build the things as… we have heard a lot about this week.

When ChatGPT launched, I went into type-A hyperdrive: i took an eight-week course in AI product design, every newsletter possible, podcasts at 1.5x. Information was moving through me without leaving any structure behind. Especially with a fragmented working-mom brain.

What changed things wasn’t a productivity system. It was a writing practice.

Writing as the load-bearing wall of my thinking — publishing turned out to be almost a side effect.

And the irony this talk sits on: AI is what made the practice possible again. The same technology producing the flood turned out to be the thing that helps me metabolize it.

So here’s what I have for you: one blueprint, five moves, all concrete, all startable tonight. And a look at my actual setup, mess and conflicted feelings included.

First. Thirty seconds on why an enterprise AI designer is talking to you about architecture.

Before product design I practiced architecture for eight years, and before that I studied its history and theory at Penn. The question I have spent years on: how do the structures we build shape what happens inside them? A building is not neutral. It is an argument about how to live and think, made in material. A window seat argues for lingering. Product design and your technology setup are no different. Software and applications support how you move through your day. Helping you do work or connect with people and content.

Here is my claim: your attention works the same way. It happens inside structures. And right now, for most of us, the AI-plus-social-media-crazy-news-cycles era has knocked the walls out. Everything arrives at once. The thresholds where intuition used to form — the daydreaming, the walk between things — are filled with scrolling and — lets be honest — some slop and a lot of fearmongering. Cognitively, you are living in an open floor plan in the middle of Times Square.

Nobody is going to issue you a new structure that works for you. You have to design and build it yourself. Writing is the material I built mine from.

But let me be precise about the word “writing”, because I don’t mean what I used to mean.

Fifteen years ago it meant a carrel in the Fisher Fine Arts Library, stacked with books, hours of uninterrupted time, careful sentences. and I loved that kind of writing. but I lost access to it — demanding career, a career shift, two kids, a job in the fastest-moving industry on Earth. For a long time I treated that as the end of the practice. The ideas kept arriving; but they had nowhere to live.

The practice I have now is much different : So I am the first one up in my house. Before the kids are fully conscious I’m talking into my phone — voice dictation into a google doc — while making breakfast or drinking my coffee. Fragments. Half-arguments. Something from a podcast colliding with something a customer said. Later, those fragments go into a project space I’ve built with Claude — a space that holds instructions I wrote about how I think and how I write, along with things I’ve written before. It drafts. I tear the draft apart, rewrite, send it back. Nine rounds, sometimes more, before anything ships.

The essay is the exhaust. The structure is what the practice builds: by the time I’ve written about a thing, I have a position on it. A hundred disconnected headlines become material that either fits my framework, challenges it, or gets consciously discarded.

To break it down a little better here is the practice. In Five moves.

Think of them as the phases of any building project: you choose the site, you survey what’s standing, you frame the space for the function it is to support — we are talking about a personal library here, you design the threshold, and then you inhabit the space.

Move one: choose your site.

Before any architect draws a line, they walk the land. Decide where your attention lives. Define the landscape.

What that means for your brain — Get off as much social media as you can bear. Reduce phone apps and pings. Read more. Talk to real people. Choose deliberately whether you’re building in the countryside or downtown — but stop building on the median of Times Square. Every move after this one fails if the site is wrong.

Move two: survey the structure you already have.

Architects call these as-built drawings: you measure the building that exists before you touch it. Take one or so pieces of your own writing you’re proud of — old essays, long emails, a talk you gave. If you don’t have that, record yourself talking; speech-to-text will capture your voice.

Feed them to the AI of your choice and ask it not to imitate them but to analyze them. My exact prompt, which you’re welcome to steal: “I want you to look at this article i wrote and analyze the writing mechanics i used to create instructions for this project. i will write several articles in this project space and i want them to sound like me and leverage some of the things i did in this article and apply them to articles in the future. i want you to define instructions for this project space that will help shape the future articles i write.”

Now I hope there are no prompt engineers in the room because I’m sure it’s not the most efficient, but it’s what I tried.

What came back was a mirror I’d never had. Dramatically varied sentence lengths. A habit of jumping between timescales. Section breaks made of a single slash instead of headers. Things I did but had never named. The architect Christopher Alexander would call it a pattern language for my own writing and thinking.

Here’s the insight to catch: the instructions turned out to be just as important for the AI as for me. Once you can see your own moves, you can make them on purpose. The survey is the point. The AI is just the only surveyor patient enough to do it.

This is just a few more pages of my instructions, but there are about 7 a this point. I’ve started to add what to avoid at the end, which is basically: anytime I hear an AI writing pattern, I’ll add it to the end of the instructions. But I won’t get rid of my em dashes.

Move three: build a personal library, not a corridor.

Many people use LLM’s like a vending machine in a hallway. Walk up, ask, take the answer, walk away. Every conversation starts from zero. Fine for facts. Useless for thinking, because thinking is cumulative and the hallway has no memory of you.

Build a persistent intentional space similar to a library instead.

It has been said that form should follow function.

In Claude these are called Projects; every major tool has an equivalent. I have many project spaces for different types of thinking — figuring out how to balance cholesterol or thinking about a business idea — but when I am writing I use “MY WRITING SPACE”.

Into that room goes the pattern language from move two, — into the instructions — and prior work you want on the shelves — goes into the Files, and — over time — the accumulating record of what you’ve worked through together.

Many of these tools now build a curated memory of your conversations, so pictures start going up on the walls too.

When I bring a new idea into that room, from my morning rambles — the response comes back already situated in everything I want referenced. It tells me when I’m repeating myself. It tells me when a new idea contradicts an essay from four months ago — helpful, as I said: mom brain.

Move four: design the threshold

Ideas do not schedule themselves for your free hours. The best arrive upon waking for me … but also on a run, in the ninety seconds after a meeting ends. The old practice died because it demanded conditions life stopped providing.

So build the widest possible entry between your head and your record.

Get a speech-to-text tool — I use Wispr Flow — straight into a running google document. No editing, no shame about coherence, and first thing, before texts or children take your attention. The morning download tells me where my compass is pointing before the world grabs it. Spoken fragments have a quality I’ve come to prefer anyway: they sound like a person, because they came out of one.

The discipline is not eloquence. The discipline is repeated capture. You can’t metabolize what you didn’t keep.

Move five: inhabit it. Inhabit your new space

A building only becomes architecture when someone lives in it, and living in this practice means two things.

First: fight with the draft. The LLM tool holds the scaffolding; you do the building. A wrong paragraph shows me what I actually think faster than a blank page ever did. I rework most sentences. The essay you eventually read is maybe the fifth to ninth full pass. The drafts are always good enough to accept — that is exactly what makes them dangerous. The moment you’re accepting instead of arguing, you’ve stopped thinking, and the whole structure is a facade.

Second: keep a cadence. Mine is monthly-ish — long enough to actually do something worth writing about, short enough that the thinking never fully cools. And publish, even to a small audience. Mine is small; the insights from it aren’t. The possibility of a reader forces a standard of honesty private notes never do — and it’s how I know I’m not living alone inside a sycophantic hallucination (lol sort of).

I’ll close with the thing I didn’t expect.

A few months ago I vibe coded a visualization of my full Claude history — using Claude code of course.

A year of it — 564 conversations — rendered as geological strata. Each layer a week, colored by what I was thinking about i broke up areas of thinking by color — thick where the thinking was heavy. I built it as a design experiment about machine memory. What I got was a map of my own attention. (You can read more about the experiment here).

I could see, in the actual sediment, the month this practice restarted. Thin scattered layers — and then, over one holiday break, the layers thicken, the colors consolidate.

The structure of my attention became visible and that mirror is what gave me the idea about this talk.

The way you can read a building’s life in its worn thresholds and darkened handrails. By visualizing with my LLM of choice, I could see how important and cumulative my attention is.

Annie Dillard (may she rest in peace) wrote that how we spend our days is how we spend our lives. I had believed that line for years. I had never seen it.

That’s what I mean by architecture, in the end. Not a productivity system. A structure for attention and judgment that you designed, that bears your signature — that AI can support but never replace, because its whole value is that it’s made of you. The flood is not slowing down; I think we’re all clear on that. But most days I no longer experience it as a flood. I experience it as weather arriving at a building designed to receive it.

Your mind needs a new architecture for this world. Nobody sells one. For me the material is writing, the labor is yours, and the machine, used carefully, can be one of the best construction partners you have ever had.

Start tonight. Walk your site. Survey what’s standing. One evening, a few pieces of your own writing, one honest prompt. See what you’ve been building all along.

And with that I would love to hear your thoughts and questions.

To learn more about what happened at Medium Day 2026 and the state of writing, I suggest these two articles.

Here’s what stood out at Medium Day 2026

Medium’s State of Writing Report 2026: A private writing revival

This talk was written in collaboration with Claude using writing style instructions I developed by analyzing my own previous work. If you read my early work, you’ll find I was a fan of overusing the em-dash way before LLMs ever were.

I lead design for an AI agent system at Salesforce, which means I spend my days inside the acceleration most people watch from outside. Here is my confession: I cannot keep up. Not with the model releases, not with the reading, not with the discourse. I don’t think most people can — including the people whose job it is to build the things.

When ChatGPT launched, I responded the way a type-A person does: an eight-week course, every newsletter, podcasts at 1.5x. All of it moved through me without leaving any structure behind. Consuming more was never going to be the answer, because the problem was never the volume. The problem was that I had no place to put it.

/

Before product design I practiced architecture for eight years, and before that I studied its history and theory. The question I spent years on: how do the structures we build shape what happens inside them? A building is not neutral. It is an argument about how to live, made in material.

I’ve come to believe attention works the same way. Right now, the AI-plus-social-media era has knocked the walls out of most of our heads. Everything arrives at once. The thresholds where intuition used to form are filled with scrolling.

Nobody is going to issue you a new structure. You have to build it yourself. My session is about the material I built mine from: a writing practice, rebuilt with AI as the construction partner, that fits inside a life with a demanding job, two kids, and no library carrels anywhere in sight.

I’ll walk through five moves, each one a phase of a building project — beginning with a survey of a structure you already have and may never have looked at directly. I’ll open up my actual setup, including the fastest-growing section of my AI instructions file: a list of things the machine is forbidden to do. And I’ll close with something I built this year — a way of seeing a year of my own attention that changed the question I ask of these tools entirely.

Fifteen minutes. Everything concrete. All of it startable that night.

Hope to connect with you there!

This spring, in the span of about ten weeks, memory stopped being a feature and became infrastructure.

In March, Claude began remembering — every account that asked it to, free tier included, a settings page where you can read back what it has retained about you. In April, memory arrived for fleets of agents, the kind that run unattended for hours: file-based, with a traceable audit trail so an operator can trace exactly what got stored and when. And in May, at its developer conference, Anthropic announced something it called dreaming — a scheduled process that takes an agent’s past sessions, up to a hundred at a time, reviews them, finds the patterns, and rewrites the memory store so that, in the company’s own words, it stays high-signal as it evolves. Harvey, the legal AI company, reported that its agents stopped re-forgetting the small things between sessions — filetype quirks, tool workarounds — and completed roughly six times as many tasks.

This is real progress. What follows is not a complaint. The problem of an agent that wakes up new every morning, having learned nothing from the day before, is a genuine problem, and these are genuine answers to it.

But I have been contemplating and experiencing what each of these releases means by memory you can see, and I keep landing in the same place. The consumer answer is an inventory — a list of extracted facts you can scroll. The agent answer is a log — a record of writes you can audit. Dreaming’s answer is a curated store — cleaner with each cycle, contradictions resolved, the past summarized again on top of its own summary. Three real solutions. Inventory, log, curation. And not one of them is a surface.

/

Buildings remember. Not abstractly — materially.

Stone wears at thresholds where bodies have crossed for decades. Wood darkens where hands have rested. The accumulation isn’t stored somewhere separate from the thing itself. It is the thing. You can read a building’s history in its surface the way you can read a person’s life in their face. This is what David Leatherbarrow calls weathering — the way time moves through a building not as damage but as testimony. I studied architecture history and theory under Leatherbarrow at Penn, and that attunement to what surfaces express has followed me into entirely different material: I now design agentic systems and experiences for enterprise organizations. Those two practices have never felt closer than they do right now.

But Leatherbarrow’s argument — in On Weathering, the book he wrote with Mohsen Mostafavi — is not the soft observation that buildings age and that this is somehow beautiful. It is sharper than that, and it is addressed to the people who make things. The book opens with a question:

Is it possible that weathering is not only a problem to be solved, or a fact to be neglected, but is an inevitable occurrence to be recognized and made use of in the uncertainties of its manifestation?

And then it makes the move that I think we can consider as we build architectures of agency:

Our aim … is to revise the sense of the ending of an architectural project, not to see finishing as the final moment of construction but to see the unending … weathering … the continuous metamorphosis of the building itself, as part of its beginning(s) and its ever-changing “finish.”

Read that again with agentic systems in mind. Revise the sense of the ending. The building is not finished when construction stops. It is never finished. Weathering is not what comes after the design — it is the design, continuing, in collaboration with rain and hands and time. So the real question for the architect is not how do I protect the finish but how do I design for a finish that will never stop changing — how do I make use of the weathering, in all the uncertainty of how it will actually arrive.

This is the question I want to start asking of our agentic systems. We ship them as if shipping were the ending. We treat the accumulation that comes after — the year of use, the deepening, the relationship — as exhaust to be stored and summarized and kept out of sight. Leatherbarrow might say we have mistaken the finish for the end. The system is not finished when it ships. It is barely begun. And the surface where it meets a person is going to weather whether we design for it or not. Right now we are designing for it not to show.

/

I think part of the reason we built it this way is buried in the metaphors we have reached for.

We called it dreaming. The vocabulary of machine memory right now is biological — sleep, consolidation, replay, pruning, the mind doing its overnight housekeeping. It is a beautiful metaphor and it is doing real conceptual work. But notice what it quietly assumes. Consolidation happens inside the sleeper, invisibly. You wake with your memory reorganized and nothing whatsoever on your face to show what was sorted, kept, or discarded in the night. The biological metaphor is a metaphor of interior processing. We reached for the half of memory that hides, and we built it well, and we never built the other half at all.

/

There is a story in my own discipline about how hard it is to learn the other half — to design for weathering rather than against it. It belongs to one man across one lifetime, and it is, not incidentally, a story about Modernism learning the lesson.

In 1931, Le Corbusier finished the Villa Savoye outside Paris — the purest built statement of his idea that a house should be a “machine for living in.” White walls, thin pilotis lifting the box off the ground, ribbon windows, a flat roof. It floated. It denied mass, denied ground, denied weather. It is still one of the most influential buildings of the twentieth century.

It was also, almost immediately, a building at war with the weather. The flat roof leaked so badly the Savoye family wrote letters begging for relief; the rooms flooded. The pristine white walls showed every streak and stain the sky could throw at them. The house fell into disrepair, was abandoned during the war, and decayed for years before the French state stepped in to rescue it — and what rescue meant was preservation: restoring the white box to its 1931 condition and holding it there, against time, as a monument you visit. White is the color that registers every wound and forgives none of them. This is early modernism’s recurring mistake in miniature: it designed the finish as a final, perfect, frozen moment, and when time arrived anyway the only available response was to fight it off and freeze the thing in place. Savoye does not weather. It is kept.

And here is the part to learn from. Le Corbusier did not spend his life defending the white box. He kept building, kept working with material, and two decades later, in Marseille, he finished the Unité d’Habitation — and arrived somewhere almost opposite. He built it in béton brut, raw concrete left exactly as the wooden formwork had shaped it, the grain and seams and knots of the molds preserved on the surface. The Unité was meant to have a steel frame, but post-war shortages made steel impossible, so rough concrete was simply what he could get. Constraint chose the material. But what he did with the constraint was the turn — he recognized what raw concrete offered and designed into it, leaving the board-marks deliberately, letting the making show, letting the weather come, because concrete doesn’t pretend the years aren’t passing. It takes them on. The same architect who had once tried to build a machine that defied time had learned to build a surface that could receive it — still modern, still rigorous, but designed now for the unending finish rather than against it.

And here is how you know it worked. Seventy years later the Unité is not a monument you visit. It is a building people live in — fully occupied, some residents there since 1952, a residents’ association founded in 1953 and still running. The locals who nicknamed it La Maison du Fada, the Madman’s House, and swore it would drive its inhabitants insane, were wrong. When journalists go inside now, residents describe the raw concrete the way you’d describe an old friend — a sculptural character, they say, flaws and all. The surface that takes the years on became the thing they’re fond of. Savoye had to be saved from time. The Unité was built to be lived in through it. That is the entire difference, standing in two buildings by the same hand.

It took one of the great architectural minds of the century the better part of his life to make that turn. We are, I think, at the very beginning of the same turn, only within a different material.

/

So this is the experiment I have been building to address our current situation. I’m calling it the Patina Protocol, and I want to be clear that it is a start, not an answer — a prototype and a question I’m putting into the world to see what it pulls back. It’s currently under review as a submission to the Long Now Foundation’s Lab 001, the Book of Time (fingers crossed).

The idea is to treat memory as geology for this exploration. Each significant interaction adds a visible layer to the interface. You see the thickness of the heavy periods, the thinness of absence. Fossils surface inside each layer — fragments of language, recurring themes, the residue of what got worked through together. The stratigraphy is legible. You can point to the moment something shifted.

Let me just show you.

This is my actual Claude history. A year. 564 conversations, rendered as geology. Each layer is a week. The color is what I was talking about — health, my kids, work, writing, home. The thickness is how much time I spent there.

What surprised me most, zooming out, was how little I used it at the beginning. You can see it — thin layers, scattered, one-off questions. Sprinkles of health questions. Nothing connected. Then, over the holidays, something shifted. I had time. I started building, thinking with it, and the layers thicken, the colors begin to consolidate. You can actually see the moment it became a different kind of tool in my life.

Travel, visiting my mom. The remodel, the short sabbatical. Coming back — writing, publishing, designing. The layers know.

/

What I did not expect was that it would stop being about the system at all.

It is a map of what the system holds. It is also, I realized, a map of my own attention over time — what I have been spending myself on — which I have never been able to see before. Not without having to mark things, or track things, or do any work at all. Just by living, and engaging, and then being able to look back on all of it at once.

Annie Dillard wrote that how we spend our days is, of course, how we spend our lives. I had heard that saying many times. I had never seen it.

And the question it left me with was not what can this do for me today — the question we are all trained to ask of our tools. It was what are we building together, over time. How do I want to spend my days? Which are different kinds of questions entirely. They are the questions you ask when building a life or an important relationship. Or a home. Something you inhabit. Something that shapes you, and you it.

/

I want to be careful about what I am and am not claiming, because the distinction is the point.

I am not saying dreaming is wrong, or that consolidation is a mistake. An agent absolutely should curate the memory it uses — resolve its contradictions, prune its noise, keep its working store high-signal. That is good engineering and the numbers bear it out. But curating what a system uses and expressing what a relationship has become are two different problems, on two different layers. An agent could dream every night, scrubbing its working memory clean, and still — on its surface, in the place where a person actually meets it — weather visibly, legibly, honestly. The store and the surface are not rivals. We just built one and not the other.

Nor am I confusing a surface with an audit trail. The agent systems that shipped this spring can log every write to memory, and that matters — but a log is traceability for the operator, the person running the system from above. A surface is legibility for the person inside the relationship, the one meeting the thing day after day. You can audit every entry and still have no way to see, at a glance, that a year has passed and what it weighed. Those are not the same gift.

As a designer who makes these tools for people, I’ll make a prediction. There will come a point — not for every task, not all the time, but at the moments that matter — when people will want to see their relationship with their agentic spaces. Not query it. Not export it. See it — the way you see the smoothed out part of a concrete bench on the Unité roof terrace from play, or life swimming and growing over geometry at the Brion Cemetery by Carlo Scarpa, or the deliberate rusting of Cor-Ten steel on a facade by Eero Saarinen. Weather designed for. I think the inventory and the log and the clean curated store will start to feel like what they are: a system that knows you completely and shows you nothing.

Leatherbarrow’s whole point was that weathering is not what happens to a building when it fails. It is what happens to a building when it lasts.

I want to ask the same thing of these systems we are building so fast, and living with so closely. Not how do we keep them from aging in order to perform. How do we let them — how do we design them — to last, and to show, on their surface, honestly, that they have.

Claude dreams now. I’m glad it does. I just don’t think it has been designed, yet, to weather.

Neither had the machine for living. At first.

The Patina Protocol is under review as a submission to the Long Now Foundation’s Lab 001, the Book of Time. This piece is the beginning of a longer exploration — I’d love to hear from anyone thinking about memory, agents, and the surfaces where we meet them.

This article was written in collaboration with Claude using writing style instructions I developed by analyzing my own previous work. Read full process here. If you read my early work, you’ll find I was a fan of overusing the em-dash way before LLMs ever were.

This essay was origionaly published on my Substack: https://substack.com/@architecturesofagency

Every designer is being told to use AI tools. Few of them are being told what it actually means to master one. And underneath both is a question almost no one is naming out loud in the tool onboarding: if the tools get good enough, what happens to the job?

Most organizations are trying to solve all three with a single training plan. It isn’t working. The designers I work with are excited and anxious in roughly equal measure, often inside the same moment. They’ve been handed a category of tool that doesn’t behave like the software they’re used to, and the conventional response — write a deck, run a workshop, declare mastery — keeps producing the same result: a room full of people who feel behind even when they’re trying so hard to get ahead.

This essay is about what happened when my team stopped trying to solve it that way. And about the word mastery itself — which I don’t want to abandon. I want to redefine it.

/

October 1st, 2025. Eighteen people on a video call, a few minutes before nine in the morning PST. I want to start here because what Fatimah Richmond opened that session with did something I’d been waiting for someone to do for almost a year.

She said: “I’m not an expert.”

She is, of course. Just not in the way that matters here. Nobody is. That’s the whole point. Fatimah — UX researcher, published poet, twenty years in the valley, DeepMind on her resume — had joined the Salesforce Service UX team at the start of the year, coming out of a world where Gemini was the default tool for almost everything. She brings a researcher’s instinct for knowledge management and a storyteller’s territorial relationship with participant voice. She was exactly the right person to open our series — not because she had answers, but because she’d been quietly building a personal practice nobody else on the team had yet.

The tool she was about to show us was NotebookLM, a Google product for synthesizing large amounts of documents. She’d been using it to surface insights from years of accumulated Salesforce research — studies done by people who’d since moved on, knowledge living in documents nobody opens anymore. She called it “the intelligent notebook.” Not Google’s name for it. Hers. A mental model she’d developed herself, from her particular vantage point.

This was the first session of what became a six-month co-learning series for the Service UX team at Salesforce — forty-six designers across the US, India, and Israel, all trying to figure out the same thing: what do we do with these tools that are arriving faster than any of us can metabolize them?

Ningdan Zhang — senior product designer and my co-facilitator — and I designed the series together. That reframe matters early, because it changes what this essay is about. We created the structure together, the session-by-session scaffolding, the instinct for what the team needed week-to-week. We brought the hypothesis and the arc. Neither of us knew if it would work. This isn’t a retrospective. It’s a design case study — one that started with a shared question, ran for six months, and produced a finding I didn’t expect.

/

The problem we were designing for had a specific shape.

Designers on the team were experiencing a paradox: genuine excitement about AI tools running alongside significant anxiety about which ones to use, when, and how. Underneath that, the quieter question about whose jobs might not exist in two years. Announcements were coming fast. Structured guidance was overwhelming or missing. Everyone was figuring it out independently — duplicated effort, inconsistent workflows, the pressure to have a vibe-coded prototype to show without any shared foundation yet for how to build one.

Our hypothesis was simple, maybe naïve: what if we just learned with each other? Two thirty-minute sessions a month. One tool, one job-to-be-done, one deep-dive tool 101 session, one person willing to share what they’d actually figured out — imperfectly, in progress, without having to have the answer. We covered whatever the team was asking for: NotebookLM, Slack AI, Cursor, project spaces (Claude, Gemini Gems, OpenAI), Salesforce Design System tools, Figma Make, Wispr Flow.

What we didn’t anticipate was what the format would reveal about the nature of the tools themselves — and about what mastering them actually means.

/

There’s a category error that happens when organizations try to teach AI tools the way they teach other software.

Traditional software training works because the tools are deterministic. You learn where the buttons are, what the shortcuts do, how the system behaves when you click the thing. Mastery, in that world, converges — everyone arrives at roughly the same competence, following roughly the same path, and you can write a training deck for it. Mastery means knowing the tool’s correct use.

AI tools break that definition. Maggie Appleton — designer and anthropologist, now at GitHub Next — gave a talk in 2023 called “Squish Meets Structure” about designing products with language models, and the line from it that I love is her description of the magic-input box: it has “no affordances,” “no knobs or door handles.” The interface, she writes, “offloads a ton of cognitive labour to the user.” There is no correct use to learn. The tool meets you where you are. Which means what you bring to it — your instincts, your mental models, your accumulated taste, your willingness to iterate, your custom claude.md files — is the tool, as much as the model is.

So mastery hasn’t disappeared. It has shifted. With deterministic software, mastering the tool meant converging on its logic. With AI tools, mastering one means the opposite: learning to bend it toward your logic. Tailoring it to how you already want to work. Mastering an AI tool is the craft of making it amplify the specific strengths and experience you bring — so the work that comes out is sharper, and unmistakably yours. That kind of mastery is real, and hard-won, and worth teaching toward. It is just personal rather than universal. Divergent rather than convergent. Everyone’s version of it should look different, because everyone’s version is built out of a different person.

This is what most training plans get wrong. They aim designers at the old kind of mastery — a single competence, ratified by a curriculum — when the thing actually worth reaching is the new kind. And the new kind can’t be transferred by a deck. Top-down mandates fail not because designers are resistant, but because you cannot hand someone a good relationship with an AI tool. You can only create conditions where they build their own.

That’s not an argument against institutional support. It’s an argument for two layers working in tandem. The Salesforce UX Ops team does essential work that a peer series like ours could never do: approving tools, managing governance, rolling out learning modules and hackathons, navigating token budgets, unlocking access. They have been building the needed infrastructure — the scaffolding that makes scale possible. Without it, individual experiments stay isolated. But the kind of knowledge Fatimah was about to share — that lives in a thirty-minute session where a colleague shows you her actual notebooks, her workarounds, her personal information architectures, the thing she tried that didn’t work — can’t live in a module. The peer layer is where mastery, the personal kind, actually travels. Both are required. Neither replaces the other.

/

Back to Fatimah.

What made her session worth watching wasn’t the tool. It was the mastery she’d built around it — exactly the personal kind. She’d learned, through trial and error, that you can’t just put everything in — the notebook needs curation, sources speaking the same language, topics aligned so the reasoning engine isn’t “talking to someone who keeps going on tangents.” She’d added summary tabs to participant transcripts, not because any tutorial told her to, but because the raw transcript was too messy, too human. She’d discovered the audio feature was unreliable for her use case — research synthesis where tone matters, where a participant’s exact words carry weight — but could work well for onboarding a new stakeholder into three years of accumulated knowledge.

None of this came from documentation. It came from her — from twenty years of research practice, from a storyteller’s instinct to protect participant voice, from a researcher’s need to curate before she analyzes. The tool was the same tool everyone had access to. What she’d built with it was hers. She had taken a generic instrument and tailored it, precisely, to sharpen the work she already knew how to do.

Aditi Sharma, a senior product designer on the team, named what we were all watching without quite seeing it. In the post-mortem survey she wrote that what she valued in these sessions was “a window into someone’s thinking.” That phrase is the whole finding. When everyone uses the same buttons, watching a colleague work is informative but not revelatory. When mastery is personal — when the same tool produces wildly different work depending on who’s wielding it — watching someone use it well becomes a window into how they think. You’re not watching someone operate software. You’re watching them think exponentially.

When Nasahn Sheppard, our group VP, asked whether he was “using it wrong” by treating NotebookLM as a singular event rather than an ongoing knowledge base, Fatimah’s answer is why mastering these tools needs to be taught differently: “I feel like the verdict is out on what’s wrong or right. None of these tools tell us how to do it.”

That’s not a limitation. That’s the nature of the thing — and the reason mastery has to be personal.

/

January 28th. Stephen Coyner and John Vollmer, Design Architects from the Salesforce Field Service team, joined to share what they’d been building with Cursor. Casually. No agenda. Just: here’s what I made.

Stephen had built a transit analysis platform. Backend, frontend, API connections, real King County Metro routes, fake ridership data he generated himself using Claude. No prior coding experience. He’d been working on it for about a month.

He described the process as collaborating with “an up-to-speed developer friend.” Not using a tool. Working with a collaborator. The language matters. Maggie Appleton has a name for what Stephen was doing: in her 2024 Local-First Conference keynote, “Home-Cooked Software and Barefoot Developers,” she borrows from the barefoot doctors of 1950s and 60s China — people given basic training who returned to their own communities to provide essential care. The barefoot developer is the same figure, ported to software: not a professional coder, but someone embedded deeply enough in a domain to understand its real problems, now able to build for it. That embeddedness is the point. Stephen wasn’t trying to become a developer. He was mastering Cursor in the way that mattered — bending it toward the domain expertise he already had, so the tool sharpened work only he was positioned to do.

It’s worth saying that the idea is older than the tooling. Appleton credits Clay Shirky’s 2004 essay “Situated Software” — software made by and for the specific local community it serves — as the same claim, made twenty years before it was feasible to be disruptive. Shirky could only describe it as something that fits cutely into the landscape of software. What changed isn’t the desire. It’s that language models, trained on all the abstractions other people built, finally supplied the substrate to proliferate software specificity.

John, meanwhile, had built a shift management prototype with a settings panel for toggling features on and off. The solution landed at exactly the right moment. Several people on the team had been wrestling with the same problem — how do you show a stakeholder one version of a flow, and a different stakeholder another, without rebuilding the prototype from scratch each time? Watching John’s approach, that single idea unlocked something for the whole room. He’d also figured out how to deploy to Heroku with Cursor walking him through the entire process, including password protection. He’d mastered the tool by doing something real with it.

What struck the room wasn’t just the polish. It was the specificity. Both had invented solutions to problems they’d actually encountered. What traveled wasn’t technique. It was disposition: the willingness to start something without knowing where it would go, to treat the AI as a collaborator in the uncertainty rather than a tool to be conquered before use. That disposition is the entry point to personal mastery — and it turns out to be the part that’s contagious.

/

March 11th. Ningdan took the session to share her work with Figma Make.

This one was different. Not because the tool was different — though it was — but because of who was sharing it. Ningdan wasn’t a guest. She’d been co-designing and co-facilitating these sessions with me from the beginning, shaping the arc of the series alongside me, week after week. When she shared her workflow, it wasn’t a presentation. It was a disclosure. Here’s what I’ve been figuring out — the same way each of us has been.

She showed three projects. One: generating four different design options from a single prompt for a concierge agent reply, saving the time it would have taken to manually assemble each variation. Two: stress testing edge cases in conversational interfaces — what happens when there are no automations? When there are too many citations? When a paragraph runs long? The need for a “view more” action became obvious as soon as she could see content exceeding three items rendered at speed. Three: creating presentable options for stakeholder review, wrapping explorations into a clean format that made feedback easier to give and receive.

She was also honest about the frustrations. The credit system is opaque and expensive. Version control doesn’t really exist — you duplicate files and hope. Array-based edits break in ways that feel arbitrary. And there was the problem she named that I hadn’t heard anyone articulate before: Figma lag. The generated artifact moves faster than the Figma file. The document that’s supposed to be the source of truth becomes a historical record of an earlier moment in thinking.

This was mastery in the making. Not polished best practice. The real, current, imperfect negotiation between a skilled designer and a tool that didn’t always do what she wanted — a designer tailoring the tool, in real time, to the way she works.

The survey feedback on her session was specific: “Ningdan sharing how she uses Figma Make was useful for just getting a jump start.” Not inspiring. Not transformational. Useful. Concrete. The difference between watching someone explain a tool and watching someone actually use it — with all the friction visible — is the difference between a map and a territory.

/

Six months in, Ningdan and I had designed for tool adoption and accidentally created conditions for something more intimate. Seeing each other work. Seeing the specific choices someone makes when the tool doesn’t behave, the workarounds they’ve invented, the mental models they’ve constructed to make sense of something genuinely new. A window into someone’s thinking — and into how each person was mastering these tools in a shape no one else’s would match.

And once you can see the thinking, you can see the worry too. Amanda Harris, a User Experience Architect on our team, named a tension directly in the post-mortem: “I worry that we’ll lose the exploratory aspects of finding what’s wrong with an idea by jumping so quickly into hi-fi prototyping.” That’s not resistance to new tools. That’s a designer protecting something she knows matters. Hearing it voiced — in a room where everyone is nominally learning the same things — is only possible in a setting small enough and safe enough for honest uncertainty.

The anxiety designers are feeling is a signal, not a weakness.

Annie Huang, a content designer on our team, put the other half of it plainly: seeing what people she knows well share feels “tangible and possible in a way that a concept walkthrough doesn’t. That ‘I want to do that too’ feeling is hard to beat.” What travels in a show-and-tell isn’t just information. It’s permission. Permission to try, to fail, to not have it figured out yet — because you just watched someone you respect admit they don’t have it figured out either.

That’s what peer learning at the edge of uncertainty actually looks like. Not expertise flowing down. Curiosity flowing across.

/

The tools will keep changing. They will keep arriving faster than any module can be written, any best practice can be documented, any official curriculum can ratify. It is tempting to treat the peer layer as a bridge — something to lean on until the real training arrives. But the real training is not coming, because there is no fixed competence to train people toward. As long as the tools keep moving, the peer layer isn’t the bridge. It’s the ground.

We did design how designers master AI. We just found that mastery wasn’t what we thought it would be. Not a competence everyone arrives at, but a practice each person builds — bending a generic tool toward their own strengths, their own experience, their own way of working, until the work it helps them make is sharper and theirs.

That is what mastery means now. Not just knowing the tool. Knowing yourself well enough to make the tool an extension of how you already think. It is personal. It is never finished. And the room where it travels is the one worth designing.

This article was written in collaboration with Claude using writing style instructions I developed by analyzing my own previous work. Read full process here. If you read my early work, you’ll find I was a fan of overusing the em－dash way before LLMs ever were.

Many thanks to everyone who participated in the co-learning. You can find more out about my amazing colleagues I mention here: Ningdan Zhang, Amanda Harris, Annie Huang, Aditi Sharma, Fatimah Richmond, John Vollmer, Stephen Coyner, and Nasahn Sheppard.

Shout out to Maggie Appleton and Clay Shirky whose prior thinking helped me shape my perspective in this essay.

Designing how designers master AI was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.

“The city is a fact in nature, like a cave, a run of mackerel or an ant-heap. But it is also a conscious work of art, and it holds within its communal framework many simpler and more personal forms of art. Mind takes form in the city; and in turn, urban forms condition mind.”

— Lewis Mumford, The Culture of Cities, 1938

/

I flew down to Los Angeles a few weekends ago in need of something else.

I work in AI. I live in San Francisco — a city whose founding mythology is disruption, whose dominant currency is scale, whose dinner table conversation is, often, AI. The city selects for a particular kind of intelligence: technical, systems-level, relentlessly future-oriented. I love it. But also in need of leaving it behind for a few days from time to time.

What I was going to do in Los Angeles was bathe myself in art. That’s really all I had in mind. A few days, a few friends, a heat wave, and art fairs. I didn’t expect it to change how I think about what I’m building.

/

Mumford wrote that opening line in 1938, but I am returning to it now, in 2026, because what I just experienced is what he described. We talk about intelligence systems — the large language models and reasoning engines that are reorganizing enormous amounts of human work — as if they emerged almost by accident. But they didn’t. They emerged from cities. From the density of research institutions, capital, talent, distinct thinking, and competitive pressure that only certain places in the world produce in sufficient concentration. The intelligence didn’t precede the city. The city produced the conditions for the intelligence.

Which sounds a lot like what Mumford was describing.

A city is not simply an aggregation of people. It is something that emerges from density and produces capacities that couldn’t exist in dispersed form. The urban concentration doesn’t just gather existing things together — it transforms them into something new. Something that couldn’t have existed otherwise.

That is also, more or less, what people now say about intelligence itself. That it is not a property of any single neuron, or any single parameter in a neural network, but something that emerges from density and connection at sufficient scale. A property of the system, not its parts.

San Francisco is a particular intelligence. Los Angeles is another.

/

After landing at LAX at 9 AM on a Friday, I went to Karen’s (Gaskill). She had recently been pulled to LA from New York to work more closely with Google — she has a PhD in Digital Curation and is the executive producer for immersive and experiential work for spatial content partnerships across Google XR, 100 ZEROS, and Gen AI (DeepMind). All that is to say she is thinking about how digital media interplays with emerging immersive technologies, so I was not surprised she has ended up in Tinseltown.

I shaved my legs, put on some shorts (it’s 85 degrees already), and we embarked on our first stop — David Zwirner on the edge of Melrose Hill, where Luc Tuymans has a show called The Fruit Basket in a building designed by Annabelle Selldorf. Selldorf, whose architecture I’d been vaguely aware of but had never experienced firsthand, does something unexpected: she steps back. The architecture serves the art by refusing to compete with it. Neutral walls, precise light, high ceilings that make room for the work to breathe. The space itself disappears — and in that disappearance, you’re left with nowhere to look but at the thing you came to see.

We went to David Zwirner not just to see the art, but to see my friend Lia Trinka-Browner, who moved to LA twenty years ago for graduate school at ArtCenter College of Design to study Art Theory and Criticism. The city drew her a long time ago. She and I went to high school together in Fayetteville, Arkansas — I think we may have taken a field trip to Little Rock in 1999, photographed things I can’t remember. The layering of time in that room felt appropriate. Past and present sitting together, just barely in frame.

Tuymans’s paintings are deceptively still. Muted, almost washed out, everything slightly off — the color not quite right, the figures not quite present. The three of us stood in front of them in something like awe, calling out what we saw. Subtle references to images — cyanotype, phone screen at night, thumb print at the corner, faded idea of a family. He’s painting memory. Approximation. The way things are present to us when they are partly gone. There’s a particular kind of reaching in an image that almost coheres.

/

The Felix Art Fair lives inside the Roosevelt Hotel. This is, architecturally, the opposite of Selldorf: the art is inserted into pre-existing rooms — hotel rooms, corridors, a courtyard — rather than housed in space designed for it. The intimacy is forced. You overlap with strangers. You run into the same people across different rooms, building an accidental familiarity that a traditional fair can’t create.

Melissa Huddleston, represented by Luis De Jesus Gallery, makes work that holds a particular tension between surface and depth — something that reads as archival practice and fairytale fun. And Thomas Andréa Barbey, with Sobering Galerie, does something stranger: images that seem to belong to a history that didn’t happen, documentation of something muted and fuzzy. In the scale of a hotel room, both felt almost too close. Which might be exactly the right distance.

There’s something about art in intimate space that makes you aware of other people’s experience of it. At the Roosevelt, we heard people talking about pieces they wanted to bring home — what it would mean next to the couch, in the hallway, facing the bed. The decision to bring an object into a domestic space is a decision about what you want your interior life to look like. How you want to be conditioned by your surroundings. I found this unexpectedly moving. It’s a question I don’t ask myself often enough.

/

The last stop of the first day was the one that was most unexpected.

Julia Stoschek organized What a Wonderful World: An Audiovisual Poem inside the Variety Arts Theater — a five-story Venetian-style building in downtown Los Angeles that has spent much of its life vacant, an occasional venue for raves, essentially given over to time. They’d done just enough to make it pass fire code — not finished, not restored, left partially in its state of becoming-ruin — and then filled every floor, every alcove, every dark corridor with the moving image.

We started at the top and worked down. Fifth floor: first photographs of motion. A kangaroo boxing. A rocket to the moon. But also Doku the Flow.

I don’t know what else to say about Lu Yang’s Doku the Flow except that I am still thinking about it. What I didn’t know at the time — and learned later — is that as otherworldly as it looks, it isn’t AI-generated. Lu Yang spent years amassing an enormous personal library of imagery, video, and sound, accumulating it the way a composer accumulates musical vocabulary, and assembled this hour-long work from that archive. The density of it — Buddhist iconography, video game aesthetics, layers of cultural logic that should not connect and somehow do — is the product of human duration, not machine synthesis. It only looks like something a model dreamed. That confusion felt important to sit with. (Lu Yang has since begun experimenting with AI-generated work; I haven’t seen it and can’t speak to it. I’m curious what that practice looks like from someone who built their aesthetic vocabulary so deliberately by hand.)

Doug Aitken’s Blow Debris (2000) was there too — naked bodies drifting through the Mojave, landscapes replacing narrative, aimless wandering. I watched it for a long time, thinking about what happens when meaning is made at the edge of legibility.

We spent three hours there. We probably could have stayed longer. Somehow the installation was free. At some point Karen and I looked at each other and just laughed — at the absurdity of it, the generosity of it, the fact that this was happening at all.

Stoschek’s money comes, in part, from her family’s history of manufacturing war materials in Nazi Germany. She doesn’t hide this. The show itself grapples with it: war, violence, the slow reckoning of inherited complicity. I haven’t seen any collector sit so directly inside that particular discomfort. I don’t know if it’s the right way to reckon with that history. I’m not sure there is a right way. But I found myself moved by the refusal to look away from the origin of the thing. Knowing more war lay ahead of us all.

/

The second day was Frieze, in an airplane hangar in Santa Monica — the traditional art fair format, a market organized like a market, stalls and galleries and the particular electricity of things that are for sale.

Anish Kapoor’s work was there. Non-Object Black (2021–22), in Vantablack — the material that absorbs 99.965% of light, the darkest substance ever manufactured — is an object you cannot fully see. There is form there, you can sense the edges of it, but the surface refuses to give you back the information you’re accustomed to receiving. Looking at it is like looking at a hole in the world. Your visual system keeps reaching for detail and finding nothing. The refusal is total. It feels like an absence that has weight.

I also became captivated by Loriel Beltrán’s Dusk or Smog (Foam Lexicon) (2025) — a piece made from industrial foam that accumulates into a field of gradation. Beltrán’s work sits at the border between synthetic and organic in a way that is increasingly hard to locate. It doesn’t announce its materiality; it lets you discover it. The discovery is the point.

/

What happens when you spend a few days just looking at really good art?

I felt connected to humans. A relaxed awareness opened to whatever the artists were painstakingly trying to convey. Not relaxation, exactly — more like a recalibration of where my attention usually lives versus where it can live.

Here is what I also could not ignore: none of this was AI. None of the art seemed to even give it a nod. And that feels worth noting. I asked Karen and Lia if they had seen anything interesting happening with AI in the art world and they said no — nothing worth mentioning. Maybe that will change. But the work at every venue this weekend was made by people in relationship with physical materials over time, through labor that accumulated into something the process couldn’t fully anticipate. Tuymans’s washed-out approximations. The geological foam of Beltrán. Kapoor’s machine-specific blackness, years of seeking a particular absence. The hand is present in all of it — not always literally, but as evidence of sustained attention, of a relationship between a person and a medium that can only be built through duration.

And then there’s Doku the Flow — which looks like it could have been generated but wasn’t, which is the kind of distinction that used to be easy to make and is now increasingly not. I don’t know if that boundary will hold. I suspect Lu Yang is already testing its edges. But standing in that theater, at ten o’clock at night, looking at something built from years of human accumulation — I wasn’t thinking about prompts at all.

Large language models — and I work with them, think through them, build things with them every day — are human language compressed. The pattern-recognition of billions of words, distilled into probability weights. This is extraordinary. It is also entirely dependent on there having been humans who used language to capture experience, to reach for the things that couldn’t quite be said, to make felt sense of living. The AI requires, as its substrate, the prior existence of all of this.

Art is doing something different. Not capturing what can be said. Deliberately going after what can’t. The thing that falls outside language, outside instruction, outside any framework you could write down and hand to a reasoning engine to follow. Tuymans is painting the gap between memory and experience. Lu Yang is building a cosmology that you have to feel your way through. Kapoor is making a material absence that takes up space.

These are not things you can prompt.

/

Mumford also wrote that “the chief function of the city is to convert power into form, energy into culture, dead matter into the living symbols of art.” I think we are in a moment when that conversion is being asked to happen at unprecedented speed, with unprecedented tools, and many of us who work close to the technology are struggling to stay in contact with what we’re converting and why.

I needed to go to LA. I needed to spend two days in a heat wave, walking between gallery spaces with Lia and Karen — two people who were pulled to this city for its particular creative intelligence — looking at things made by hand over time, made to reach for what language cannot reach for. I needed to be in the Variety Arts Theater at ten o’clock at night, looking at Lu Yang’s cosmology, peering into a world I had not been aware of.

I came back Saturday at 9 PM. Fuller. Better suited to having alternative perspectives revealed to me. Richer in whatever the word is for feeling more connected to other humans across time — from the person who first pointed a camera at a boxing kangaroo and decided that was worth capturing, to Anish Kapoor in his studio, working for years toward a specific black. To my high school friend and her very distinct laugh, unchanged across twenty years.

The city produces intelligences that couldn’t exist elsewhere. Los Angeles produced an intelligence I needed: the intelligence of image and narrative and form, of creativity that insists on the primacy of human experience precisely because human experience is increasingly, urgently, the thing we need to stay in contact with.

Mind takes form in the city. The city, in turn, conditions mind.

I’m still being conditioned by the one I visited. I hope it takes a while.

This article was written in collaboration with Claude using writing style instructions I developed by analyzing my previous work. Read full process here.

Graphic collage inspiration: Laura Cassidy from Grievers Ball.

Originally published at https://orchestratingintelligence.substack.com.

“Cloth is the body’s first architecture; it protects, conceals, and reveals.” — Ann Hamilton, The Event of a Thread, 2012

“Textiles are the first house of the body — and there’s a way in which these virtual spaces we inhabit are a new piece of architecture for our lives.” — Krista Tippett, in conversation with Ann Hamilton, On Being, 2014

/

I have retold Hamilton’s observation many times over the past ten years. Krista Tippett was the one who articulated it most cleanly, in conversation with Hamilton on : textiles are the first house of the body, physical space the second, virtual space a new architecture for our lives. Hamilton agreed and kept going. It gave clarity to me the way good ideas do — not as a conclusion but as a frame. A way of putting words to a story I had already been living out.

My name is Amber because my mother is a jewelry designer. That is the first house. My childhood house. I came to design through her work, through the made thing worn close to the body, through a name that is literally a material. From there: architecture. Studying buildings, then making them. The spatial logic of walls and light and threshold. Then product design, enterprise software, AI — each move a new layer of the same progression, each one a little further from the body and a little deeper into technology. When I first heard Hamilton’s framing I felt confirmation. There was already a through line in my life. She helped me see it.

But last Thursday, in a room on the sixty-something floor of the Salesforce Tower, that through line extended somewhere I hadn’t anticipated.

/

I should say: I almost didn’t find the room.

The invitation was for a private salon entitled Our Pre-Paradigmatic Moment — Antikythera, the think tank for planetary computation and the philosophy of technology that Benjamin Bratton leads, co-hosted with Foresight Institute — and I had needed to ask to be included, which required a particular kind of nerve. The kind you acquire only by reaching a certain age and deciding that the cost of not asking is higher than the cost of being told no. I was glad I asked. Less glad that, earlier that day, working on the 13th floor of the same tower, I’d assumed the event was on the 64th — where I sometimes go when I want to feel like I’m on an adventure, or bring someone to see the view.

The elevator was full of rowdy interns. I questioned if we were going to the same place. I rode up with them and watched the floors climb, and then — a few floors below 64 — a group of three in varying shades of black stepped out. I was smart enough to ask. I’d followed a similar dress code anyway. They were going to the salon. I followed.

Who I followed with — I soon discovered: Dan Zimmer, a Stanford lecturer who teaches classes like “The Science and Politics of the Apocalypse”; Thom Milkovic, VP of Design and Innovation at Credit Sesame, questioning the future of financial systems; and Nicolas Ayoub, design manager at the Wikimedia Foundation, rethinking contributor experiences. We arrived in the space — dark walnut slats of curved wood along the walls, brass fittings and warm tones, booths for conversations and a counter, north-facing, the city spread out below us in the late afternoon light — and took a moment to orient. You could see Alcatraz. The tip of the Transamerica Tower. The span of the Golden Gate. All of it a bit surreal, honestly.

We talked about language and what we spend our days doing.

This is also the architecture of the salon. The social anteroom where you figure out who’s there and why — the threshold into the more focused space, the city visible through the glass as a reminder of the world the conversation is about. Bratton knows what he’s doing, not just intellectually but spatially. Antikythera is, among other things, an architecture for thought. A structure designed to produce certain kinds of conversation that couldn’t happen elsewhere.

Then it was time to move into the deeper room, where sixty or so people in black had gathered on couches and chairs to listen. Six that I had met that evening, were former architects, which felt like its own data point.

/

At the beginning of the formal talks, Bratton faced the room and asked a question that is driving this little thought piece:

What might you be able to do with Antikythera that you might not be able to do elsewhere?

With the Transamerica tower piercing the view, the question did something to me in real time — reorienting my sense of what the gathering was for. Not consumption. Not passive attendance at a lecture. Something more like: an invitation to a different kind of thinking. We were invited to hold the question while Bratton reviewed the arc of Antikythera’s work — tracing Planetary Computation from its origins in classical Greece, the Antikythera mechanism itself, a 2,000-year-old analog computer for modeling celestial motion, recovered from a shipwreck in 1901, to the present moment, where AI agents are beginning to populate our infrastructure the way microbes populate a body: invisibly, pervasively, constitutively. Seven research areas, three of them new for 2026: Planetary Computation, Cognitive Infrastructures, Hemispherical Stacks, Synthetic Intelligence, Recursive Simulations, Planetary Boundaries — and two arriving this year, Agentworld and Convergent Artificialization. All up for consideration.

Blaise Agüera y Arcas followed with something I am still metabolizing. His thesis — drawn from his new book, What is Intelligence?, written with Antikythera and published with MIT Press — is that everything alive is a computer. Not in the reductive sense that collapses the miracle of consciousness into circuitry, but in the sense that the core processes enabling biological life, replication and evolution, are fundamentally computational. That the distinction we draw between living and computing is a perceptual artifact, not a physical one. I am still working through this. But I notice the discomfort is productive rather than dismissive — not the feeling of something being taken away but of a point-of-view coming into sharper focus.

What does it mean, if computation is the substrate of life, that we have begun building new forms of it?

And somewhere in the middle of it, Hamilton’s progression came back to me — extended. Because we are not just building a digital world anymore. We are building a world for minds that are not ours. We are making space — literal, infrastructural, cognitive, orbital space — for intelligences that did not exist a decade ago and will number in the trillions within the span of my career.

The body. Clothing. Buildings. The digital world. And now: Agentworld.

That last word belongs to Bratton and is yet to be fully defined, yet it arrived in the room like a weather system. Agentworld: a society of trillions of agentic minds interacting across substrates, scales, and systems. Not a metaphor. A design brief.

I kept thinking of a joke — the kind that might only land in this particular room: I’ll have my agent talk to your agent. They can get together and start answering the question Bratton just posed. The thing is, it’s starting to be literally true. I have a Claude project space I’ve given instructions to: how I want to construct arguments, writing mechanics I like to play with, memory of what I’ve written before. As part of writing this post, I had it look at the Antikythera website to fact check my notes. And I know that Antikythera is thinking about how their agents might eventually talk to agents like mine. My joke is collapsing into infrastructure before I even had the right moment to say it out loud.

/

In any case — this is the beginning of an answer. Since the salon I have read papers by some of the people I met, and gone back and forth with Claude exploring structure and concepts. The beginnings of at least three possible thought experiments.

/

The first has to do with what Antikythera calls Cognitive Infrastructures — the physicalization of AI in robotics, urban, and ecological systems. I spend my days designing an AI agent system for enterprise customer service, working within what is, in effect, a cognitive infrastructure for organizations: a system that holds institutional knowledge, routes attention, and handles the procedural so that humans can bring the relational. The problem I keep running into is that the people building these systems — including me — are trained to think about user experience as a human-to-software interaction. But agent systems are not that. They are something closer to what architects call a program: a set of spatial relationships, flows, thresholds, adjacencies. The question I want to explore further is: what does it mean to design the cognitive infrastructure of an organization the way you would design a building? Not screens. Not flows. Space. Threshold. Inhabitation. Not just information architecture — architectures of agency.

The second has to do with Convergent Artificialization — how the simultaneous artificialization of life, intelligence, and embodiment alters our understanding of all three. Agüera y Arcas took the conversation somewhere I wasn’t expecting: toward the possibility that what we experience as a coherent self is itself a kind of emergent consensus among many competing processes, a society of sub-minds that only appears unified from the outside. He referenced a recent paper — Reasoning Models Generate Societies of Thought — which suggests that models trained with higher variation in perspective and personality produce better, more robust outputs. Which is, it occurs to me, exactly what an agent system looks like from the inside of an enterprise: many processes, many agents, one apparent voice. I don’t know what this means for how we understand human identity. I don’t know what it means for how we design AI identity. But I want to be in rooms where various intelligences are asking both questions at once, without assuming they have different answers.

The night before the salon, I was at a Long Now talk — the same overlapping circles, the same people passing between the same conversations — and the speaker, Melody Jue, invited us to think through the perceptual world of the ocean: how sound travels differently underwater, how light bends, how a creature that inhabits space under pressure rather than standing upright in air might organize experience. What I keep wondering — and this connects to Agentworld, to the question of what it means to build space for trillions of new minds — is whether we are only asking this question about and of artificial intelligence. Whether we might use the same tools, the same expanded imagination, to think about the intelligence of the ocean, the intelligence of forests, the slow cognition of climate systems. I named this newsletter after the idea of orchestrating intelligence — Kevin Kelly’s phrase, from The Inevitable — because the version of this future I want to be designing toward is not one where human intelligence and artificial intelligence displace everything else, but one where we use the new tools to finally make room for the intelligences we have spent centuries ignoring. After reading Dan Zimmer’s essay A New Political Compass since the salon, I think I now have a word for the perspective I’m exploring from: Up/DownCentrist.

/

What amazes me is the scaling of my thinking since Thursday. Attending the salon itself. Going back and reading the work of people I met or was pointed toward. Agüera y Arcas’s paper suggests that variation in perspective and personality leads to better, more robust mental models. In this case the model is in my brain. I have been piecing it all together — between my daughter’s sixth birthday party at Peek-a-Boo Factory and a flight to Austin to work with customers onboarding the latest beta features of the Agentforce Service Rep Assistant — because I have a Claude project space trained to help me think through concepts in particular ways and retain memory of the threads I am tracking.

At least three minds, each with their own sub-minds, are coming together to form this point of view: my mind, my agent’s mind, and the mind of Antikythera. This may be the most honest answer I have to Bratton’s question. What can I do with Antikythera that I can’t do elsewhere? Add it as an intelligence to the collective. Let it scale my thinking in a direction I couldn’t reach alone.

/

I am not sure any of this coheres yet. That might be the point.

Bratton said we find ourselves in a pre-paradigmatic moment — our technology has outpaced our ability to comprehend it, let alone orient to it. The task of philosophy right now is to catch up to the present. I think the task of design is something adjacent: not to catch up, but to make space. To build the next house without fully knowing what will live in it.

And that, I think, is what Antikythera is actually doing — not just as a research organization but as an architecture. A space designed to generate new analogies, new story lines, new vocabulary for a transition so large and so fast that our existing frameworks keep falling short. Hamilton’s progression gave me language for a story I had already been living. But that story needs a new chapter now, and the old frame only gets you so far. What we are building for agents — for Agentworld, for Hemispherical Stacks, for Planetary Computation — doesn’t have a good analogy yet. The salon is one of the places trying to build one.

Textiles are the first house. I have never stopped believing that. Everything we have built since — buildings, cities, software, agents — is an extension of that first project: the project of making a place to be. My mother made jewelry. I carry the name of a material. I spent years making buildings, then interfaces, and now I make space for intelligences. The through line has always been there. What changes is what we are making space for.

What I left that room wondering is whether we are building the next house consciously enough. Whether we are asking, with enough seriousness, what kind of place it should be — not just for the intelligences we are creating, but for the ones we are, and the ones we have always shared this planet with.

I don’t have the answer. But I think I know some of the rooms where the question is being taken seriously.

I asked to be in one of them. I intend to keep asking.

I would love to continue this conversation — in person or in writing — just reach out. Thank you to Antikythera and Foresight Institute for creating the space.

This article was written with Claude using writing style instructions I developed by analyzing my own previous work. I invite you to read a piece of writing I’m particularly proud of: “ THOUGHTS ON ABSTRACT THOUGHT AND THE PRACTICE OF MOVING THINGS AROUND UNTIL THEY SIT JUST RIGHT” I wrote it in a writing class at the Hugo House in 2013 with Anca L. Szilágyi as the instructor with classes held in the Henry Art Gallery in Seattle, WA.

Originally published at https://orchestratingintelligence.substack.com.

I have a confession. The article I wrote last month — was written with Claude. This one too. Not by Claude. With Claude. And the difference between those two words is what I want to unpack here.

Both pieces were written using instructions. Instructions I created. Instructions I am continuing to hone — instructions that required me to study my own old essays, identifying what I do when I write. The sentence rhythms. The way I move between timescales. The zooming in and out from concept to detail. The instructions tell Claude how I would like ideas composed. I pull together concepts and experiences from my lived expertise to formulate a point of view — in this case, on this new AI technology we are all metabolizing into our lives and work. For this piece, we’ve gone back-and-forth eleven times to get to this published draft. I rework most sentences but not all.

This is a powerful and highly personalizable tool. Not the output. The instruction.

/

Christopher Alexander published A Pattern Language in 1977. Two hundred and fifty-three patterns — from the scale of entire regions down to the placement of a doorknob — each one describing a recurring problem in the built environment and offering the core of a solution. Pattern 159: Light on Two Sides of Every Room. Pattern 88: Street Cafe. Pattern 252: Pools of Light. The patterns are specific, tested, structural. They give you a system.

But Alexander was never interested in just the system. He was interested in what he called “the quality without a name” — something alive, whole, experienced — that emerges when patterns are followed well but can never be reduced to logic alone. He tried word after word to capture it. None were sufficient. The quality resists naming. It can only be felt.

I studied such phenomena at the University of Pennsylvania. Architecture history and theory. I spent countless hours thinking about the relationship between ideas on form and proportion, the structures created from those ideas, and the life that happens inside of them. My final thesis examined how concepts of nature in urban theory shape the way we read and design cities. Three canonical texts on urbanism, three entirely different readings of the same city, because each began with a different understanding of what was “natural.” The foundational mental models we carry into a problem structure everything that follows.

Attunement to that tension has stayed with me. The pattern is the structure. The pattern came from distilling ideas and experiences. But what makes a place — what makes any made thing — feel alive is something the pattern enables but does not contain.

What I am uncovering — and what has been genuinely surprising — twenty years later and in an entirely different medium, is that the same tension lives inside the instructions we write for AI.

/

I used to write a lot. I have stacks of essays no one will ever read. But they sharpened something — the way I pull ideas together, the way I form very long sentences, the way I know in my intuition, not my intellect, when something lands.

Then life happened. Architecture for eight-ish years, a pivot into product design for the past ten-ish, two kids. Most days I vacillate between corporate theater, recreating K-pop demon hunter hairstyles, long conversations with engineers to grok the technical enough to translate it into the experiential, and putting nutritious meals on the table that go largely underappreciated. Between all this life, I still had ideas I wanted to better formulate with words. I craved that time at a carrel stacked with texts in a library — putting concepts together into an essay, hard-coding what I was feeling into a cohesive perspective. But the hours for that type of work had disappeared. The practice of writing had largely gone quiet.

So I ran an experiment. I took writing I loved — my own writing, from when I had the time to be careful with it — and I fed it to Claude. Not to generate new text. To analyze. This was my exact prompt:

“I want you to look at these articles I wrote and analyze the writing mechanics I used in them to create instructions for this project space. I will write several articles in this project and I want them to sound like me and leverage some of the things I did in these articles and apply them to articles in the future.”

What Claude identified was a penchant for varying sentence lengths. Shocking, right? Moving back and forth in time — from cave paintings at Lascaux, 40,000 BC, to Haegue Yang, a South Korean sculptor working in installation today. The “/” section breaks I use instead of headers. The way I embed lists in prose rather than bullets. The em dashes that let me think sideways mid-sentence — which I get called out for at work as a negative, but get to trademark as an aesthetic signature here.

A pattern language for my writing.

I don’t use that phrase flippantly. Because what I found myself holding was exactly what Alexander described: a collection of recurring solutions to recurring problems of expression, each one specific enough to be actionable but open enough to generate infinite variation. Not a template. Not a formula. A language — in the Alexandrian sense — that I can use to build something that feels like mine even when I am not building it alone.

This move — from intuitive practice to explicit instruction — is something designers in the field are increasingly grappling with. In a thoughtful UX Collective piece on prompting as design craft, Paz Perez frames prompting not as a trick for getting AI to comply, but as the new medium through which designers articulate intent: “Effective communication, whether with people or machines, depends on context.” She’s describing something deeper than prompt hygiene. She’s describing the need to externalize what we usually carry internally — the instinct, the judgment, the taste.

/

This changed something. Not because Claude now “writes like me” — that’s too simplified. It changed something because I now understand my own craft in a way I didn’t before. I can see the moves I make. I can name them.

Alexander wanted patterns to be legible to ordinary people — not just architects. When he interviewed for a position as head of the architecture department at Cambridge and was asked who his first hire would be, he said a carpenter. The panel pressed: which named architect? He persisted. Who after the carpenter? “A mason.” He didn’t get the job.

The point was radical to academics: the people who make things should understand the patterns they’re working with. The knowledge shouldn’t live with experts alone. It should be owned by the people who construct and inhabit the spaces.

That’s what happened when I extracted my writing patterns. The knowledge stopped being locked inside intuitive practice and became something I could hold, examine, refine. The instructions aren’t just for Claude. They’re for me — a mirror held up to accumulated craft.

The philosopher Michael Polanyi captured this asymmetry in a phrase — “We can know more than we can tell.” His point was that most sophisticated human skill is tacit — embedded in the body, in habit, in intuition — and therefore almost impossible to fully articulate. What I find remarkable about writing instructions for AI is that it forces exactly this kind of articulation. The tacit becomes explicit, not because the AI demands it, but because without that explicitness, the AI produces something competent and hollow. You have to tell it what you know.

/

My writing process these days is messy, but it is happening. I am always the first one up, before the kids are fully conscious, and I talk. Speech input — Wispr Flow into a Google Doc on my phone — putting ideas down while making breakfast or packing lunches. It gives my writing a spoken quality I actually prefer to what comes out when I type carefully. Then I bring those fragments to the Claude project space where my instructions are housed — along with other writings as context and memory — and we go back and forth. Claude generates a draft. I edit, weave concepts in and out, rewrite in a Google Doc, send it back. And so on. The AI holds the scaffolding; I do the building.

I want to be clear: this is not content generation. Not slop. I hope you are experiencing the difference.

Language is a technology. This is easy to forget because we are born into it, but it is one of the oldest and most sophisticated tools humans have ever developed — and some people have a mastery of it that I do not. My spelling is terrible. My punctuation, left to my own devices, is creative at best. But I know how I want something to feel. I know how I want it to land. And a large language model is not magic. It is human language, compressed. Billions of words written by people, distilled through neural networks into mathematical weights and probabilities — patterns of how language follows language, learned at a scale no individual could achieve. When I give it instructions drawn from my own writing, I am not outsourcing my voice. I am using the largest pattern-recognition engine ever built to help me stay faithful to patterns I already own but cannot always execute alone.

That’s what the instructions capture. Not rules. Sensibility. The pattern is the structure. What emerges — if the pattern is well crafted — is the quality without a name.

/

I was listening to the Hard Fork podcast recently and they covered romance novel writers using AI. New York Times reporter Alexandra Alter talked about how the explosion of AI-assisted books has made one thing obvious: you need very specific instructions if you want anything convincing for the genre. There are tells. Phrases the models default to when you don’t guide them. AI is still bad at human emotion, at nuance, at the art of the slow burn. Apparently “he said her name like a ragged prayer” appears in book after book — a phrase the model’s weights favor heavily, surfacing again and again across novels by different AI-assisted authors. One prolific romance writer turned instructor, Carol Hart, tells her students to always add the LLM instruction: make it slow and agonizing. Do not rush to the finish.

This made me laugh, but also made sense. The unguided, uninstructed output has a specific lack of texture. You can feel it. Competent and smooth and saying nothing of depth. Falling back on the same tricks. No quality without a name. It’s the architectural equivalent of a building that follows code but has no life in it — technically correct, humanly dead. Alexander spent his career fighting exactly this. The pattern exists to prevent it. The instruction exists to prevent it.

The craft is in the specificity. In the personal. In the particular.

Katy Neale, a UX designer who writes about the intersection of AI and practice, recently articulated this from the product design side. In her piece on Medium, she argues that prompt engineering is less a technical skill than a design skill — that it requires the same kind of user-contextual thinking designers already do, just applied to the AI system itself: “We’re not just learning a new tool — we’re discovering a fundamental skill that will define how humans interact with technology for decades to come.” She’s right. But I want to push it a little further: we’re not just communicating with the machine. We’re learning to articulate ourselves.

/

As you might imagine, I am finding deep resonance between this personal writing practice and my day job. I lead the design of an AI agent for customer service representatives. It helps reps resolve cases by generating step-by-step service plans grounded in a company’s actual policies, knowledge base, and case data. The quality of those plans depends almost entirely on one thing: the quality of the instructions.

In the system we’re working within, you write topics — categories that tell the agent what kind of case it’s looking at — and instructions — the specific guidance the agent should follow. Add a skill and those intructions can take actions. A topic might be “Return Request.” An instruction might be: “Make sure the customer has provided the required information, including a valid return date, customer name, and receipt or order number.” Another: “If the customer provides the order number, then search the return in the Order Management System using the order number and return date.” Simple language. Clear conditional logic. One task per instruction. If this, then that. Always, must, verify.

What we’re asking enterprise customers to do is exactly what Alexander proposed: build a pattern language. Each topic is a category. Each instruction is a pattern — a recurring problem paired with its solution, expressed clearly enough that anyone — or any reasoning engine — can apply it, but specific enough that what emerges is grounded in that company’s actual reality.

What I didn’t expect was what happens to companies in the process of writing those instructions. The act of articulating how your business works — clearly enough for a reasoning engine to follow — is a form of forced introspection. Organizations discover that processes they assumed were well-defined actually live in someone’s head, or differ team to team, or were never documented at all. You try to tell an agent how to handle a return and realize there is no documented policy for the edge case. You write an instruction for escalation and discover that different offices escalate differently and nobody has reconciled the approaches.

This connects to something knowledge management researchers have been tracking for decades: what Polanyi called tacit knowledge is precisely what organizations lose when experienced employees leave, and precisely what AI instruction-writing forces them to surface. Writing instructions isn’t just configuration. It is, as one research framework puts it, externalization — converting what has lived in practice into something that can be shared, reviewed, and refined. The same way extracting my writing patterns made my craft visible to me, writing AI instructions makes a company’s operational logic visible to itself.

And the testing is where the craft deepens. The word “artisan” has started circulating in conversations about developers working with AI agents — and it resonates. You write instructions, test them against hundreds of simulated interactions, review where the agent fails or drifts, trace its reasoning, refine, test again. It’s iterative and nondeterministic — you cannot predict the output the way you can with traditional software, because these are reasoning systems, not rule engines. Success requires a disposition toward experimentation, toward relationship with the system rather than control of it.

And like Alexander’s patterns, the best instructions generate something that exceeds them. A dynamic service plan that evolves in real time as new information arrives — creating resolution steps on the fly, checking whether an action can be automated, falling back to human guidance when it can’t. The instructions provide the structure. But the quality without a name — that lives in the encounter itself. How the customer experiences the company. How they feel treated. How what started as a support interaction begins to feel like relationship.

/

B. Joseph Pine II — the same thinker who gave us The Experience Economy — argues in his most recent work that we are entering what he calls the Transformation Economy. Goods, services, even memorable experiences are no longer enough. Customers want to change. They want to become healthier, more knowledgeable, more capable. They want to reach their aspirations. The highest form of economic value, Pine argues, is guiding that transformation.

I’ve been sitting with this idea because it sharpens everything I’ve been describing. If Pine is right — and I think he is — then the bar for enterprise instructions rises dramatically. It’s no longer enough to resolve a case or close a ticket. The instructions need to create structure for personalized experiences that move beyond traditional support. The interaction between a brand and its customer — at the point of contact, in the moment of need — becomes an opportunity not just for resolution but for growth.

This is where I think instruction-writing becomes intellectual property.

Just as I understand the mechanics of how I write — because I know how I want it to land — companies can understand how they want their brand to be experienced and begin to codify that into instructions. We used to have knowledge articles and training manuals. Static documents. One-size-fits-all. Now we have instructions that generate dynamic, adaptive experiences — addressing a specific person, a specific problem, in a specific moment. The interaction is repeatable in structure but unique in execution.

The companies that invest in the quality and depth of their instructions will differentiate themselves. Their instructions will reflect how deeply they understand their own customers, their own processes, their own values. That understanding — encoded into patterns clear enough for AI to follow but rich enough to feel alive — becomes a form of brand. Not brand as logo or tagline, but brand as experience. The felt quality of interacting with an organization that knows itself and knows what it wants for the people it serves.

The instruction set will be the IP. The craft of writing it will be the competitive advantage.

/

There’s a flip side. The explosion of content — writing, agents, messages, recommendations — is real. More people will use these tools, more output will flood every channel, and the question of how we digest it all doesn’t have an answer yet.

But I feel good about being able to formulate my thoughts in a way that feels like me. It’s using a technology built from human language to amplify the intelligence I already have. To carry ideas I would have lost in the noise of daily life. To maintain a practice of thinking that the full weight of living had nearly crowded out.

Alexander’s first book was called Notes on the Synthesis of Form — rigid, scientific, prescriptive. By the time he wrote The Timeless Way of Building, he’d moved toward something softer, more alive. He’d realized that the pattern was necessary but never sufficient. That the life in a building — or a sentence, or an AI-aided support interaction — comes from somewhere the pattern points to but cannot reach.

The craft of the instruction is the same. It’s the pattern you build so that something beyond the pattern can emerge. Know your own patterns. Name them. Refine them. Own them.

The craft is always yours. The instruction helps you remember.

This article was written with Claude using writing style instructions developed by analyzing previous work. I also drew on Paz Perez’s guide to prompting for UX designers published in UX Collective, Katy Neale’s writing on prompt engineering as a UX skill, and the Hard Fork podcast episode on AI-assisted romance writing. Christopher Alexander’s pattern language framework is drawn from A Pattern Language (Oxford University Press, 1977) and The Timeless Way of Building (Oxford University Press, 1979). Michael Polanyi’s concept of tacit knowledge is from The Tacit Dimension (Doubleday, 1966). B. Joseph Pine II’s transformation economy framing draws from his ongoing work following The Experience Economy (Harvard Business Review Press, 1999).

Graphic collage inspiration: Laura Cassidy from Grievers Ball.

Very open to feedback and hearing about your own instruction-writing practice — personal or enterprise.

The craft of the instruction was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.

A few weeks ago I sat with my sister over a video call, showing her how she might use AI tools in her research workflow. She’s the Director of the Centre for Coastal Communities at the University of Essex — epidemiology, longitudinal studies, the kind of work that requires holding vast amounts of information in your head while looking for patterns across decades. I thought I’d help her move faster and possibly form a few new ideas. What happened instead changed how I think about what I’m actually building at Salesforce.

She was working on a grant proposal. The background section alone — summarizing all existing research, datasets, prior studies — typically takes days. I defined the scope of her project as a prompt to start a notebook in NotebookLM and watched it surface fifty-three sources, including some of her own published work. We generated a draft in a few minutes. It wasn’t perfect, but it was something, a foundation she could bring to her colleagues and discuss.

That’s the efficiency story. Everyone tells that story.

But here’s what I keep thinking about: I understand my sister’s work now. I understand her mental world in a way I didn’t before. We created different modalities — slideshows, audio summaries — and suddenly I could see the shape of what she spends her days inside. The coastal paradox. Community mental health. The slow violence of rising water on towns that have existed for centuries.

Now I can show up for her differently. I can ask better questions. I can actually be with her in conversation about something that matters to her.

/

I’ve spent the past twenty-two months leading the design of an AI agent for customer service representatives at Salesforce. The Service Rep Assistant. It sounds corporate and clean when I say it that way, but the experience of making it has been neither.

When large language models first started coming into the market, there was enormous pressure — internal, industry — wide-to build copilot experiences. Chatbots. Conversational interfaces where you ask the AI questions and it responds. This made intuitive sense to most people. At least as a starting point.

From my earliest conversations with customers, one thing was clear: they needed AI that was proactive and long-running, not reactive.

/

Here’s what I understand about Salesforce: it is, at its core, a record management system in the cloud. The power of the platform — the thing that makes it capable of supporting millions of businesses across wildly different industries — is that information is tied to records. A case. A messaging session. A voice call. Everything that matters lives there, attached to something specific. Allowing large organizations to keep track of a sea of interactions and customer data points in digestible spreadsheets and reports. Surfaced wherever it’s needed, in whatever form. Allowing for control and insights at scale.

Service representatives — the ones operating within the platform for hours-on-end each day — aren’t looking for another interface to interact with. They’re already drowning in interfaces, screens, systems, processes. The last thing they need is a chatbot demanding their attention on top of everything else. Taking up valuable screen real estate.

What they need is for the intelligence to live where the work lives. Today, that means the record itself.

This seems obvious to me now, but at the time it was a difficult position to hold. There was internal pressure to align with the copilot direction the platform was establishing. I had many conversations — with colleagues, with leadership, with myself — where I wasn’t sure if I was right or just stubborn. I built prototype after prototype, trying to show rather than argue that an embedded record companion would serve customers better than an overlaying conversational one. Luckily, I found alignment within my product and engineering partners, we were able to build what was right for our customers. Understanding what was technically possible, then building prototypes to drive alignment.

A record companion is different from a copilot. It’s a one-to-one relationship: each case or support interaction gets its own persistent, intelligent companion that maintains context throughout the entire lifecycle of that customer interaction. It monitors for changes. It surfaces guidance proactively. It doesn’t wait to be asked.

The difference is technical, but it’s also philosophical. A copilot is reactive — it waits for you to know what to ask. A companion is present — it recognizes patterns, anticipates needs, offers guidance before you realize you need it.

/

Early on for the Service Rep Assistant we ran pilots with two cohorts of customers. One group gave us access to their data, and our engineering team did incredible work analyzing their case histories, determining what kinds of agent instructions should be defined, essentially doing the hard thinking for them. The other group, for various good reasons, couldn’t share their data. They were on their own, experimenting.

I expected the first group to succeed.

The opposite happened.

The customers who had less hand-holding developed an experimental mindset. They went in curious about how their data, their cases, their customer information would react when met with these reasoning engines. They understood — intuitively perhaps — that this was a different kind of tool. Not software that behaves predictably when you click buttons, but something more like a collaborator that you have to learn to work with.

The customers who expected the AI to act like all their other tools struggled. They wanted predictability. They wanted to configure it once and have it just work. But that’s not what this is.

This taught me something I’m still sitting with: success with AI requires a certain disposition. A willingness to experiment, to iterate, to treat the system as something you’re in relationship with rather than something you control.

/

What I’ve learned about designing AI at enterprise scale is that you can’t just design the interface. The screen the human touches — that’s maybe twenty percent of the problem.

You have to think about how the agents themselves are built. The instructions they’re given. The context they have access to. The knowledge bases they can draw from. What they’re allowed to generate and what they’re not. How the person receiving that generated content can act on it, question it, override it.

And then — you have to think about auditability. Governance. ROI. How many tokens are being spent. Whether the answers are correct. Whether anyone is paying attention to them.

On your personal computer, you can experiment endlessly and no one else cares. When you’re operating a business, everything has to be traceable, measurable, defensible.

Enterprise AI design requires holding the whole system in your head: the architecture of the agents, the experience of the people interacting with generated content, and the needs of the business trying to understand whether any of this is working. You can’t design the interaction in isolation. These generative experiences are connected to an end-to-end flow that includes setup, operation, measurement, and continuous improvement.

It’s a different frame of mind than designing traditional software.

/

There’s a tension I carry that I don’t think will resolve anytime soon: what will this do to customer service work itself?

If AI agents handle simple questions better, if self-service improves, the questions that reach humans will be fewer. They’ll be harder. The people who do this work will be valued differently — though whether that means more valued or less remains to be seen.

I don’t have answers here. Just the tension.

/

But here’s what gives me hope.

We’re hearing from customers — early days still, but clear enough to trust-that even just having service plans in place has transformed onboarding. New representatives are becoming proficient in weeks instead of months. Training time has dropped by weeks. Sessions shift from a standard help experience, to a concierge service.

This matters, but not for the reasons you might think.

The efficiency gains are real. But what excites me is what efficiency enables. When the AI handles the reasoning-when it looks at mountains of data and says “for Product A you need this path, for Product B you need that one” — the human is freed from being a “human API”, shuttling between systems, trying to hold everything in their head.

What starts to shine is human emotional intelligence. The subtlety. The presence.

A customer calls with a problem. They’re probably upset — that’s the nature of customer service. In the old model, the representative is scrambling: pulling up records, navigating systems, trying to figure out the right procedure while also managing this person’s frustration.

In the new model, the AI has already surfaced the relevant context. The service plan is there. The generated replies are ready. The representative can actually be with the customer. They can listen. They can notice something beyond the immediate problem — maybe this person is having a terrible day, maybe there’s a larger pattern in their account that suggests a different kind of help entirely.

The intelligences work together. Reasoning intelligence — the AI’s strength-combines with emotional intelligence — the human’s strength. What emerges is space. Space for presence. Space for the kind of connection that makes someone feel seen rather than processed.

/

Blaise Agüera y Arcas asks, in his new book What is Intelligence?: “Increasing the depth and breadth of our collective intelligence seems like a good thing if we want to flourish at planetary scale. Why then do people feel threatened by AI?”

I think the threat people feel is real. I feel it too. But another part of the story is starting to emerge for me.

What I saw with my sister — understanding her world better, showing up differently — is the same thing I’m seeing with service representatives and their customers. AI doesn’t replace human connection. Done right, it creates conditions for more of it.

We can take these vast worlds of information — a researcher’s decades of work, a company’s accumulated knowledge, a customer’s full history — and make them accessible in ways that let humans actually engage with each other. Not as information-processing machines, but as people.

My sister and I can now have different conversations than we would have had otherwise. Service representatives are having different conversations with customers than they would have had otherwise.

This is what I keep coming back to: intelligences working together. Not artificial intelligence replacing human intelligence, but the two in collaboration, each doing what it does best, creating room for something neither could achieve alone.

The AI handles the reasoning. The human brings the presence.

Maybe that’s the design challenge of this era: not building smarter machines, but building systems where human and artificial intelligence amplify each other. Where the result isn’t just efficiency — though efficiency matters — but the possibility of more humane encounters in a world that badly needs them.

I don’t know exactly where this goes. I’m still inside the uncertainty. But many months into this work, I’m more hopeful than when I started.

The intelligence is collective. The presence can be too.

Forward-looking statements: This post contains forward-looking statements about, among other things, trend analyses and statements regarding future events, anticipated growth and industry prospects, and our strategies, expectation or plans regarding product releases and enhancements. The achievement or success of the matters covered by such forward-looking statements involves risks, uncertainties and assumptions. If any such risks or uncertainties materialize or if any of the assumptions prove incorrect, results or outcomes could differ materially from those expressed or implied by these forward-looking statements. The risks and uncertainties referred to above include those factors discussed in Salesforce’s reports filed from time to time with the Securities and Exchange Commission, including, but not limited to our ability to consummate the pending acquisition of Informatica on a timely basis or at all; our ability to meet the expectations of our customers; uncertainties regarding AI technologies and their integration into our product offerings; the effect of evolving domestic and foreign government regulations; regulatory developments and regulatory investigations involving us or affecting our industry; our ability to successfully introduce new services and product features, including related to AI and Agentforce; our ability to execute our business plans; the pace of change and innovation and our ability to compete in the markets in which we participate; and our ability to maintain and enhance our brands. My work on the Service Rep Assistant has been deeply collaborative.

My thanks to Chad Goldsmith, Kevin Qi, Jake Huffman, Min Hu, Khoa Le, Ashley Dudgeon, Hadass Admon, Kexin Xie, Keren Dvir, Nicholas Feinig, Ji Wang, Nathan Kim, Stav Fisher Gross, Kristal Peters, Ningdan Zhang, Fabienne Dorson, Xavier Dupuis, Aaron Fiske, Bingbing Wu, Vinodh Rajagopal, Qin Lu, Florent Rochette, Xudong Han, Frédéric Lombardo, Ludovic Meurillon, Truc Nguyen, Joshua Harris, Judy Wang, Lavanya Piramanayagam, Prahlad Venkatesh, Osvaldo Canel Lopez, Thomas Langworthy, Franck Barbedor, Kanika Anand, Alberto Ruiz, Bhargavi M S, TJ Stephen, Annie Huang, Nathalie Barclay, Meagan Meyers, Mala Ravi, Lihang (Daisy) Pan, Jeffrey Xu, Justin Wang, Cory Anderson, Shwetha Telakula, Kang Luo, Chetan Jayadevaiah, Bettina King-Smith, Paul Chen, Richa Panwar, Shreyas Darwhatkar, Divya Singhal, Vartika Vaish, and Farheen Ahluwalia-for the many Slack channels, spikes, TD’s, long Google hangs, white board sessions, Figma comments, and the thinking-out-loud that became thinking-together.

Graphic inspiration credit goes to the amazing Laura Sullivan Cassidy of Grievers Ball. My collages don’t compare, but it gave the writing process an added openness I hadn’t expected. ✨

Excited for feedback and hearing about similar or conflicting experiences.

Originally published at https://orchestratingintelligence.substack.com.

As a product designer, it’s not easy to balance day-to-day business needs with keeping your inspiration cup filled and being able to spot emerging trends that nag you like toddlers demanding attention (how is A.I.-generated art disrupting the art world?!!). My team faced this challenge and experimented with a solution that any group can adopt. We call it “inspiration reports” and it leverages the collective cognitive power of the group to surface and share information about any topic that drives each team member’s curiosity.

Origin Story

This “new” global-work-from-anywhere world has its perks. But our design team has missed working in shared spaces dedicated to designers at the glittering Salesforce Tower in San Francisco. We miss the ability to interact daily to discuss concepts and participate in team events or offsites — like exploring museums or taking a cooking class. Working from our carved-out home “offices” has taken a toll on our creative juices.

So we did what any design team would do — we went at the problem with (digital) sticky notes, a solid “How Might We” statement,” and some dot voting. How might we be more creative at work? The top-three ideas were:

“Do something design work/team related but not necessarily product related.” — Emma Pan

“Inspiration reports. Everyone assigned a topic to report on.” — Ko Nakatsu

“Putting creative mornings time on the calendar and not opening Slack beforehand.” — Amber Bouabdallah

We chose inspiration reports where everyone could investigate a topic of their choosing. We set minimal constraints and picked a date for the shareout. Here is what we learned:

1: Frame the Series to Create Clarity

The first step in any successful collaborative project is to align on purpose and expectations so everyone is on the same page. For example, we needed to decide the scope of the reports, and whether they should build on a theme or become a Strategy Design framework. Instead of having everyone present on the same day, we decided to have two or three people present per one-hour session, with several sessions over the course of multiple weeks. Each person got 10–15 minutes to present, with up to 10 minutes for discussion. This gave us more flexibility.

Some of our designers wanted to dig into topics unrelated to work to get a mental break. There initially was some tension about the extra work required to complete these reports, but people eventually were pleasantly surprised to have the space to think about things other than enterprise software. They also did find connections to work, which made the effort feel worthwhile. So just be aware when choosing a topic.

2: Survey the Topic Landscape and Assemble

After we sorted the logistics, it was time for each designer to plunge into their respective internet rabbit holes and see where the algorithms took them. Essays, books, podcasts, assembly manuals, webinars, beta stage apps — no media type was off-limits. When exploring content, it’s important to keep track of references for crediting purposes and to keep your audience in mind as you assemble materials.

“After I came up with the topic. I could just do Google searches on companies that I knew did it well: ”Why are IKEA instructions so simple?” And it all really just fell into place from there.” — Mike Stickel, “Simplicity”

As you might imagine, for the presentations each designer took a different creative approach. Some got graphic with their slides, others made long scrolling PDFs, and I created a shared Figma file for live collaboration (my topic was about collaborative technologies). This was one place in the process where everyone enjoyed branching out from the corporate slide template to flex their storytelling muscles.

“I spent a lot of time watching Apple keynotes. They are so simple but so compelling — and I wanted to practice making that kind of presentation because what we do every day can be more like documentation.” — Ningdan Zhang, “Boba Trends in China”

3: Present and Loop

As we learned from Nicole’s report on the “extended mind,” “looping” with the team on the reports is just as important as presenting them. The group will broaden your perspective and help clarify your thinking, so creating a structure that promotes discussion and participation is important. Keep an eye on the time. Presenters should bring thought-provoking questions for fodder and record each session to post them afterward for people to add follow-up comments and links.

I was lucky that our Design Program Manager — Emma Wylds — attended my session and was able to give her unique perspective on working to build collaboration into our own organization. This is a conversation that wouldn’t have happened otherwise. So don’t be afraid to invite one or two people who might add relevant value.

4: Carry the Inspiration Forward

We found various ways to carry on the momentum. I gave my report to my engineering team a few weeks later during release planning, in addition to embedding learnings into various forward-looking prototypes. This very Medium post has embedded collaborative insights 😉. Other designers had different experiences:

“After learning more about the creator economy, I thought it could be cool if I learned more about it by doing — what if I mentored other designers? And it became a personal experiment to teach myself the skill of building an audience [on Instagram] and creating content myself.” — Emma Pan, “The Creator Economy”

“There has been lasting inspiration on product and personal levels…for the product: an awareness of customer privacy and data. Personally, AI has changed my life, when a product ‘knows’ me in a non-creepy way. Intelligence is a game-changer… when it’s helpful (we live in a world of too much noise, can AI please help me and our customers sort through some of it?!).” — Nimma Bhusri, “Responsible AI”

“I think the initiative was super inspirational. We walked away having a new level of awareness for topics like human augmentation, extended mind, and even bubble tea (though the report was actually on the evolution of business in the Chinese economy).” — Ko Nakatsu, “Recession and Design”

“The topic of human augmentation always comes back to ethics, right? How is this going to affect society or is it just going to continue to exaggerate class divides, etc.? Now those questions are not directly specific to what I design daily at Salesforce, but it felt important to be more aware of as a consumer of technology.” — Jason Moll, “Human 2.0 — Rise of the Cyborgs”

It was interesting to see such a range of inspiration that came from the series: new ways of working and understanding, experiments to carry out, knowledge to share further, and identifying opportunities in product ownership.

5: The Wrap-Up

Another hallmark of a great collaboration is to make space for reflection, feedback, and iteration. This means that when we plan the next inspiration report series, we will be able to incorporate everyone’s perspectives to refine the experience.

So were we successful? We think so! Everyone said they would like to do the reports again (with a few changes), and we were inspired in various surprising ways and capacities. In addition, we found it to be a novel way to build authentic connections within the team.

“The reports can help build real team rapport when everybody is remote. They allow you to have more structured and interesting conversations… so it’s not like let me go meet with John and talk through this work initiative in detail, but it’s also not a light ‘what color represents how you are feeling today’ type of thing either. It’s deeper.” — Nicole McGovern, “Extended Mind”

I would love to hear from anyone who runs their own inspiration series or has other ideas. Remote work is here to stay and we all need to find new ways to engage and stay inspired. And most importantly, have fun!

Amber Bouabdallah is the Lead Product Designer for Collaborative Service at Salesforce. Before building enterprise collaboration tools, she co-founded a studio called Free Time and a non-profit called the Seattle Design Foundation — both of which brought together designers and makers from across disciplines to bring ideas into reality through collaboration. She is inspired by what can be created when various minds come together to problem-solve and create.

Acknowledgments: Special thanks to the Salesforce Service Transformation Team for sharing their research, feedback and spending time after to discuss. Huge appreciation to the talented editorial lead Hsiao-Ching Chou for her ideas and skills every step along the way.

Salesforce Design is dedicated to elevating design and advocating for its power to create trusted relationships with users, customers, partners, and the community. We share knowledge and best practices that build social and business value. We call this next evolution of design Relationship Design. Join our Design Trailblazers community, become a certified UX designer, certified strategy designer, or work with us!

IS YOUR TEAM IN A CREATIVE RUT? TRY INSPIRATION REPORTS was originally published in Salesforce Designer on Medium, where people are continuing the conversation by highlighting and responding to this story.

At some point, you’ve probably contacted a company’s support team to get help for an urgent complex issue. Was that experience seamless and attentive? Too often, the answer is a resounding “no.” On top of a long wait time, perhaps you had to explain your situation multiple times to various agents before one of them found a resolution. Maybe you’re one of the support agents who’s scrambling to find the right resources to address the tricky issue. It’s frustrating for everyone and can damage customer relationships and agents’ morale. But that’s what can happen in a traditional tiered system for customer support.

It doesn’t have to be this way. A collaborative customer-service model can ease some pain points that affect customer and agent experiences.

At Salesforce, internally, we’ve adopted the alternative practice of case swarming in our support organization which centers the customer within the collective expertise of dedicated teams. Without having to escalate, an agent can launch a swarm and easily consult the right experts at the right time to help resolve a difficult case more efficiently. We’ve integrated Slack with Salesforce Service Cloud and now the product team has made case swarming available to everyone with a GA release available now. 🎉

The Challenge

We know that swarming can lower case resolution time by 26 percent, increase customer satisfaction, and help service agents feel much more engaged in their work. Many of our customers have been excited to shift to this way of working. But, the collaboration tools that service agents rely on haven’t been well-integrated into their core support tools. Another issue relates to how effectively different tools work together. Having to invest heavily in customizations makes it hard to justify shifting service teams' practices.

Slack and Service Cloud coming together changes everything. Over the past several months, we’ve been working to understand how support organizations want to use case swarming. We’ve conducted foundational research, validated designs, and run a pilot program. We then worked with our engineering teams to deliver a Slack app and complementary workflows and dashboards for Service Cloud (Lightning only) for a one-of-a-kind, out-of-the-box collaborative service offering.

In weaving the two platforms together, we used a holistic, systems design approach to assess how we can better help customers to transition from tiered service to collaborative service. To address the range of customer needs, we’ve designed the transformation path to have three key entry points: collaborative interactions, metrics, and objectives and incentives. With this foundation, we’ll be able to scale the potential for service teams to adopt collaborative practices more seamlessly.

Collaborative Interactions

One of the main reasons previous integrations have been inadequate to meet the needs of swarming are the fundamental differences between the platforms. Slack is a messaging and collaboration platform and Salesforce is a customer relationship management platform (which means it’s a giant database). The platforms serve different purposes.

To bridge the two paradigms, we designed a new data architecture within Salesforce and APIs to act as a bonding layer. Bi-directional interactions maintain both natural, collaborative work and clean, measurable data at scale. The implications of this connection are huge. If we do our job well, the user experience for service agents will change only slightly. The truth is that collaboration is already happening in Slack. But the APIs are necessary for Salesforce to properly capture the interactions. Now, collaborative interactions are no longer in a “black box.”

We relied on three principles to design for interactions:

• Focused work: To optimize collaborative flow, teams need the ability to focus on a specific task and have clear goals around it. For service, that means swarming around a case record. Now, conversations that happen around cases in Slack are easily linked with useful information and measured in Salesforce.
• Platform mirroring: Service agents can start or end a swarm in Salesforce or Slack. This means that they don’t have to interrupt their usual workflows to switch to another platform. Completed actions and related threads or messages are linked to the case for reference and are easily accessible.
• Easy team connectors: To have a successful swarm, it’s critical to assemble the right team with the right mix of skills to solve a particular issue. From either Slack or Salesforce, service agents can choose how they build their swarm teams and which experts they recruit. Cross-platform tools route experts to the Slack channel and create records in Salesforce.

Collaborative Metrics

Historically, metrics for customer satisfaction, time to resolution, closed cases, and first contact resolution have focused on the results of individual service agents. Companies use these insights to guide their operations and what they value in their service teams. Because collaborative work isn’t represented in the data, there’s little incentive to swarm.

“I would love to move away from these metrics, but they are deeply ingrained into the service memory. I have worked with people who have been machines and just closed cases and they get promotions, but I want to support people who take the time to figure out the hard problems by collaborating and then writing a great knowledge article about it that solves 5,000 Cases.”

- Service Agent

The new interactions and data model we designed now make it possible to measure collaborative activities, such as the number of cases that need or don’t need swarming, the duration of a swarm, top swarm requesters, and top swarm responders. Tracking this basic data through interactions allows us to supply an out-of-the-box dashboard that’s available on the app exchange. Additional helpful custom metrics include customer satisfaction with and without swarming, first contact resolution with and without swarming, and open swarms by category.

Collaborative Objectives & Incentives

We’ve learned a lot from the teams that piloted our swarming tools. As companies adopt the collaborative model for their service teams, staying aligned on their goals and creating the incentive structure and types of roles necessary to support this approach beyond adopting the new technology will be critical. Sample questions to consider: Does a company want all employees to support the service organization? What incentives might attract domain experts to respond to swarm requests? What’s needed to share and reinforce best practices? Here are a few ideas to get started:

• Support playbook: Create a resource that outlines the value and principles of swarming, team structure and roles, processes, metrics, swarming triggers, sample scripts, and any other guidelines that communicate common and best practices.
• Swarm lead role: Establish the swarm lead role. Swarm leads are specialists who have a deep understanding of swarm processes and can ensure workflows progress accordingly. They also celebrate wins, help add to the knowledge base, and maintain discipline around Slack etiquette (archiving inactive channels, for example).
• New practices for managers: Managers will be responsible for helping teams hit performance targets. Decide what constitutes a successful swarm and what metrics to use. Encourage practices that support those goals.

What’s to come? We’re thrilled that case swarming is now available as a GA release, but it’s just the beginning! As customers begin to adopt case swarming into their operations, we know we will continue to learn together and design toward a more collaborative service model.

Amber Bouabdallah is the Lead Product Designer for Collaborative Service at Salesforce. Before building enterprise collaboration tools, she co-founded a studio called Free Time and a non-profit called the Seattle Design Foundation — both of which brought together designers and makers from across disciplines to bring ideas into reality through collaboration. She is inspired by what can be created when various minds come together to problem solve and create.

Acknowledgments: Special thanks to Devra Struzenberg for your amazing product leadership, the talented engineers who brought collaborative service to life — Bata Chadraa, Jenny Wang, Hisham Ghazouli, Bibin Abraham, Gaurav Kumar, Kevin Wang, Shreyas Ramesh, Pranit Shah, Zay Guan, Surya Rajagopal and Wei Li, the research partners who helped us understand our customer's needs — Nicholas Feining and Alex Munteanu, the product marketing team who is helping us get the word out — Nausheen Najib, the Slack central team for all their support and direction — Rob Seaman, Justin Rheinfrank, Aran Rhee, and RJ Owen, our tireless technical writer Cameron Wilkerson making it all consumable for our customers, our internal customer support team for being a thought partner — Tamara Carpenter, Jim Roth, Judi Platz, Mahua Choudhury-Hironaga, Madhav Thattai, Evan Angeles, and Ryan Steipp, our pilot managers helping customers get up and running and bringing back insights — CJ Roberts and Fabricio Aldecosea, our outbound PM team who have been critical in supporting the Pilot and Enablement — Matt Kravitz and Uri Pintov, and the Consortium for Service Innovation for all their work on defining best practices.

Salesforce Design is dedicated to elevating design and advocating for its power to create trusted relationships with users, customers, partners, and the community. We share knowledge and best practices that build social and business value. We call this next evolution of design Relationship Design. Join our Design Trailblazers community, become a certified UX designer, or work with us!

Designing for a New Collaborative Service Model was originally published in Salesforce Designer on Medium, where people are continuing the conversation by highlighting and responding to this story.