NotebookLM just got a major update, and it’s better than ever.

It’s still a research tool, but the new features in the latest update — Gemini 3.5, the Deep Research function, and Chat Artifact creation — mean there are now more possibilities for new use cases.

It’s not just a document summarizer anymore. It’s more like a full research workspace. It can run code now, too. It can find sources, generate finished outputs, spreadsheets, slide decks, charts, and more.

If you haven’t reviewed how you use it, you might be missing out a lot of time on the table.

Here are seven use cases worth building into your actual workflow as repeatable systems.

1. Brief Yourself Before a High-Stakes Meeting

The problem with most pre-meeting prep is that you end up reading the same documents you’ve already read, looking for the thing you know is in there somewhere. Now NotebookLM changes that dynamic entirely.

You can drop your relevant documents for a meeting: the deck you’re reviewing, the proposal you’re responding to, or the contract you’re negotiating.

Then ask it direct questions, like what are the three biggest risks in this proposal? Or what commitments did we make in the previous contract that apply here?

It then gives specific, cited answers in minutes instead of you spending an hour re-reading the whole thing. As someone dealing with plenty of documents, this saves me plenty of time.

2. Turn Competitor Materials Into a Structured Analysis

Competitive research used to mean spending a morning reading through websites, press releases, case studies, and earnings calls, before we’d manually build a comparison in a spreadsheet. Now, NotebookLM’s new data table feature can do most of that work.

All it takes is to upload the competitor’s public materials as sources. Their website, their latest whitepaper, their product pages, any earnings call transcripts you have access to. Then ask NotebookLM to extract and compare: pricing structure, positioning, product features, customer focus, key claims. It builds a structured table you can export directly to Google Sheets.

The update’s source discovery feature helps here too. If you start with a few known competitors and want to find more, you can let NotebookLM search the web for related materials and add them with your approval. The whole research pass that used to take half a day now takes under an hour.

3. Get Up to Speed on a New Industry or Regulation Fast

Every professional hits a moment where they need to become conversant in something unfamiliar, fast. A new client in an industry you haven’t worked in before. A regulation that just passed and now affects your work. A market you’re evaluating for the first time.

The old way was to read everything you could find and hope you retained enough to sound credible. The NotebookLM way is to build a dedicated notebook: upload the regulation text, relevant industry reports, analyst commentary, whatever public materials you can gather. Then interrogate it like a tutor. What are the five things I need to understand first? What are the common points of confusion in this regulation? What would a practitioner in this space say I’m missing?

Generate an audio overview for the commute. Come back and drill specific sections. By the time you’re in the room with someone who’s lived in this space for years, you can hold the conversation.

4. Work Through a Reading Backlog That’s Been Sitting There for Months

Everyone has one. The pile of reports, articles, papers, and decks that were supposed to be important but never got read because the week got away from you. Most of them still matter. You just don’t have the bandwidth to get through them linearly.

Drop the whole backlog into a notebook. Not all of it, but the specific cluster that relates to a project or question you’re working on right now. Then ask for a synthesis. What are the main arguments across these pieces? Where do they agree? Where do they conflict? What are the two or three things I should understand from all of this before I make a decision on this topic?

You stop reading to accumulate and start reading to decide. That’s a meaningfully different relationship with the information you’re supposed to be on top of.

5. Turn Client Documents Into a Queryable Knowledge Base

Consultants and agency owners spend an unreasonable amount of time hunting for things they’ve already seen. The brief from six months ago that had the relevant budget breakdown. The client feedback from Q2 that explains why the current brief is shaped the way it is. The contract clause someone referenced in a meeting last week.

Build a notebook per client or per engagement. Upload everything: briefs, meeting notes, contracts, emails you’ve exported, past deliverables. Then instead of searching through folders, you query the notebook. It finds the specific thing, cites the source, and shows you exactly where to look.

The update’s team sharing feature means you’re not the only one with access. Your whole team can query the same source of truth, which cuts down on the “do you know where that document is” conversations that eat more time than anyone tracks.

6. Generate a First Draft of Your Presentation or Report

This one sounds too good but it works within reasonable limits. Upload the sources behind a presentation you need to build: the research, the data, the previous version of the deck, any reference materials. Tell NotebookLM what narrative you need and who the audience is. Generate a slide deck.

What you get back is not the final deck. The slides are images, not editable elements, so you’ll export to PowerPoint and rebuild from the narrative it proposes. The value isn’t in the output itself, it’s in cutting the brainstorming phase. You stop staring at a blank deck wondering what the story should be and start responding to a proposed structure you can agree or disagree with. That shift from blank page to reaction mode saves more time than it looks like it should.

The new code execution feature adds another layer for anyone working with data. If your report needs to show trends from a dataset, you can give NotebookLM the data and ask it to run the analysis and generate the chart directly. No export, no separate script, no manual build.

7. Onboard a New Hire Without Rebuilding Context From Scratch

Every time a new person joins a team, someone senior has to spend hours downloading institutional knowledge that lives in their head or scattered across old documents. NotebookLM gives you a way to build that transfer systematically instead of reactively.

Build an onboarding notebook. Pull in everything that matters: product documentation, past client work, internal guides, relevant strategy documents, the post-mortems from the projects that shaped how the team operates. Structure it as a resource the new hire can query themselves.

Instead of a one-hour orientation where someone talks and the new hire takes notes, they can spend their first week interrogating the notebook, asking it what they don’t understand, generating a study guide from the internal materials, even creating an audio overview of the team’s history to absorb on the way in. The senior person is still available for questions. But the questions get better because the new hire arrives with actual context.

Where it still falls short

NotebookLM is a retrieval tool, and that it only responds to questions you already knew to ask is the limitation itself.

It does not show you the relationships between your sources, the gaps in what you know, or the connections that exist in the material you uploaded but never occurred to you to query.

In other words, you get answers, but do not get a map.

That is the gap I built Constella to fill. It takes the same sources you would upload to NotebookLM and maps them visually, pulling documents, notes, and integrations into a canvas where relationships surface on their own. The insight you were not looking for shows up in the structure of the graph before you think to ask for it. If that layer is what you’re missing, it is worth trying.

Start with one notebook this week. Pick the use case that matches where you’re losing the most time. That’s the only instruction worth following here.

7 NotebookLM Use Cases for Knowledge Workers (Updated for the 2026 Release) was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.

Every time AI gets better at something, someone writes the obituary for the whole category.

The second brain industry is next in line. If AI can organize your knowledge, surface what you need, and connect your ideas automatically, why pay for a dedicated app?

But that question assumes the wrong thing. AI is not replacing these tools. It is changing what they have to be. And that change is more significant than most people in this space are willing to admit.

1. The value shifted

For most of this industry’s history, storage was the product. You built your vault, captured your notes, linked your ideas. The app that gave you the cleanest way to store things won.

That’s over.

The value now lives in retrieval. It’s not just about the search anymore, but the ability to surface the right piece of knowledge at the right moment, without you having to go looking for it. An app that stores everything beautifully but retrieves poorly is now a liability, not an asset.

The race has changed, and most legacy tools haven’t caught up to that yet.

2. Context is the new moat

The best second brain app used to be the one with the best interface or the most integrations. But that’s no longer true.

The app that wins is the one that knows you best. Your patterns, your unfinished thoughts, the connections between ideas you never made explicitly. The one that understands not just what you saved, but what you actually meant by it.

That’s context. Building deep context is a fundamentally harder problem than building a good note editor. The apps that solve it will be very difficult to displace. The ones that don’t will start to feel like fancy filing cabinets.

3. The stickiness changed

The more of your intellectual life an app touches, the harder it becomes to leave.

The new lock-in in this industry is not pricing or proprietary file formats. It’s that the app has learned how you think, and moving means starting that process over from scratch.

That kind of stickiness didn’t exist five years ago. It’s becoming the most important competitive variable in the space.

4. The habit problem is gone

Second brain apps used to have a dirty secret. They only worked if you were religious about using them.

You had to tag consistently, link deliberately, and review regularly. If you fell off the habit for two weeks, your vault became a mess. These apps were powerful for the disciplined. For everyone else, they became a graveyard of half-finished notes.

AI removes that dependency. The app has to work even when you’re inconsistent. It has to surface things you captured three months ago and forgot about, make connections you didn’t make yourself, and stay useful through the gaps. If it can’t do that, it’s not built for how most people actually work.

5. The expert bar dropped

For a long time, getting real value from a second brain app meant becoming a student of methodology. Zettelkasten, PARA, evergreen notes. There were books, courses, YouTube channels, entire communities built around doing this correctly.

That world is shrinking fast.

AI handles enough of the underlying complexity that you no longer need a system to get value from these tools. You can be inconsistent, undisciplined, and still walk away with something useful. The community that built a religion around this space did real work. But AI is quietly making most of the doctrine unnecessary.

That’s good for the market. It opens the category to people who were never going to build a Zettelkasten. But it also means the apps that built their identity around power users have to rethink who they’re building for.

6. The competitive map changed

Three years ago, building a credible second brain app meant a large team, significant engineering resources, and years of product iteration. The established players had a real edge.

That’s no longer the case.

AI compressed the development timeline dramatically. A small team can now build something that competes directly with tools that have been around for a decade. The moat shifted from scale and resources to intelligence and depth. What matters now is how well your AI understands the user, not how large your team is.

This is reshaping who can play in this space. The incumbents are uncomfortable about it in ways they’re not fully saying out loud.

The apps that figure this out first won’t necessarily be the biggest ones. They’ll be the ones that understood the product was never the notes. It was always the thinking behind them.

This was the problem I couldn’t stop thinking about when I started building Constella. Most tools store what you give them and wait. I wanted something that reads across everything you’ve saved, notices the connection you missed between two things you captured three months apart, and tells you exactly where your thinking has a gap before you think to ask.

That’s the whole point.

AI Will Not Replace Second Brain Apps. But It Will Change Them Significantly. Here’s How. was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.

When you’re a solo founder, you’re — literally, and figuratively — the institutional memory of your business. You make every decision, every freelancer brief, and every lesson learned from a bad hire sits inside your head and nowhere else. And the moment you need to recall any of it, you realize the entire operation is running on something that evaporates every time you sleep.

Most productivity advice tells you the answer is a better note-taking system. But here’s the thing. Notes might be useful to some extent, but they do not surface connections. They do not tell you that the contractor you are about to rehire has a pattern you documented three projects ago.

When Andrej Karpathy published his LLM wiki pattern earlier this year, most people read it as a tool for personal knowledge management. A smarter second brain. But I see it differently; it’s actually something a solo founder like me actually needs: a business brain, aplace to run the operation from, not a place to store what you are learning.

I built one over the last few months.

The Difference Between a Second Brain and a Business Brain

A second brain is optimized for you. It helps you retain ideas, connect concepts, and build on what you have read. The unit of value is your personal intellectual growth. Roam, Obsidian, Notion; every PKM tool on the market is built around this model. You are the beneficiary.

A business brain, on the other hand, is optimized for the operation. The unit of value is how fast someone — you, a freelancer, a future hire — can get to the context they need to make a good decision or do good work. The beneficiary is the business itself.

Most solo founders who pick up Karpathy’s pattern apply it the first way. They feed it articles and papers and newsletters and build a beautiful wiki of ideas they find interesting.

That is a fine use of it.

But it misses the more immediate problem. If you are running a business with a rotating cast of contractors and no team to distribute institutional memory across, what you actually need is a system that knows your business better than anyone you have ever hired.

The difference is about what you put in.

What a Business Brain Actually Holds

Mine has six categories.

The first is freelancer briefing context. Every time I bring someone new in, I used to spend hours reconstructing the same background: what we are building, who the audience is, what the voice sounds like, what has been tried before, what failed. That context now lives in the wiki. A new contractor queries it and gets a brief that would have taken me an afternoon to write from memory.

The second is a decision log. Every significant call I have made — a pricing change, a positioning shift, a hire, a product cut — goes in with the reasoning behind it. This one has saved me more than once. The issue is rarely that I forgot I made the decision. It is that I forgot why. The decision log gives the reasoning back.

The third is contractor notes. These are patterns across engagements, not performance reviews. What kind of feedback does this person need to do their best work? What did the last three projects with them reveal? Where does the collaboration break down? The wiki builds a profile over time that is more useful than any formal evaluation I have ever written.

Product thinking is the fourth. User interviews, feature debates, things I heard in a sales call that I was not ready to act on. The raw material is scattered across Slack, email, voice memos, and notes from calls. The wiki pulls it together so that when I am writing a spec, I am querying a real history rather than guessing at what users said six months ago.

Hiring criteria is the fifth. After enough hires, you start to see what actually predicts a good outcome for your specific business. The wiki builds that picture incrementally. Before I post a role now, I query it. What comes back is more accurate than my gut at this point, and my gut is not bad.

The sixth is personal working rules. The operational principles I have figured out the hard way — how I work best, what conditions push me toward bad calls, what I keep underestimating. This one is personal, but it belongs in the business brain because the business runs on me.

How I Set Mine Up

The structure follows Karpathy’s pattern closely. I have written about the full setup. The short version: raw sources stay in one folder, untouched and immutable. The LLM compiles them into a wiki folder of structured markdown pages — one page per concept, person, project, or decision.

What changed for the business use case is what counts as a raw source. For a personal wiki, raw sources are articles and papers. For a business brain, raw sources are meeting notes, Slack exports, call transcripts, old briefs, client emails, and anything else that captures what actually happened in the business. I use the Obsidian Web Clipper to convert anything browser-based to markdown. Voice memos get transcribed first. Everything else goes in directly.

The discipline is feeding it consistently. A decision that does not go in is a decision the wiki cannot learn from. I treat the log like a commit now. Something happens, I document it while it is fresh, I drop it into raw. The wiki updates on the next compile.

How I Actually Use It

Three scenarios cover most of what I do with it day to day.

When I onboard a new freelancer, I have them start with a query session rather than a brief I wrote. They ask the wiki questions about the product, the audience, the voice, the history of decisions. What comes back is grounded in everything I have fed it — not a summary I wrote once and never updated. The freelancer gets a more accurate picture than I could reconstruct from memory, and the onboarding conversation becomes a refinement rather than a data dump.

When I am about to make a significant hire or rehire decision, I query the contractor notes section before I say yes. The patterns are in there — the ones I would talk myself out of in a real-time conversation but cannot ignore when they are written out plainly. I wrote about one version of this lesson in a piece on a $100k bad hire. The wiki is how I avoid writing another one.

When I am writing a product brief, I query the product thinking section for everything relevant to the feature or change I am working on. User feedback I logged months ago surfaces. Decisions I made and reversed come back with their reasoning. The brief I produce is grounded in a real history rather than whatever I can recall on a given afternoon.

The compounding effect is the part that took me longest to appreciate. Every query I run makes the wiki more useful because I tend to add context after the session — corrections, updates, new decisions. The business brain gets more accurate over time, not less. That is the opposite of how human memory works, which is the point.

What Changed

Before I built this, the business ran on memory that existed only in my head. Every new freelancer started from zero. Every old decision had to be reconstructed. Every hiring call was informed by gut instinct dressed up as pattern recognition.

What kills solo founder operations quietly is rarely bad product or bad marketing. It is the gradual erosion of context that happens when one person is carrying everything and there is no system for any of it to land in. The knowledge problem is invisible until it costs you something.

The wiki fixed the infrastructure. But I still needed to see across it — not just query it point by point, but understand the shape of what I knew, where the connections were, where the gaps were. That need is what pushed me to build Constella. I wanted a canvas where the compiled knowledge became navigable visually, where I could see how a product decision connected to a hiring pattern connected to a contractor’s feedback, without having to ask the right question first.

The wiki is the memory. Constella is how I see across it.

FAQ

Do you need to know how to code to build a business brain this way?

No. Karpathy’s pattern works with Claude.ai or any agent that can read files. Claude Code makes the workflow faster and more automated, but it is not required to get started. The folder structure is the important part, not the tooling.

How long before it becomes genuinely useful?

The initial structure takes an afternoon. The wiki becomes useful after a few weeks of consistent feeding — typically once you have ingested enough raw sources that the compiled pages start surfacing things you did not remember logging. That moment tends to arrive faster than you expect.

What if my raw sources are messy? Slack threads, voice memos, email chains?

Feed them in anyway. The LLM is good at extracting signal from unstructured input. A messy source is more useful than nothing. The more important habit is getting material in at all, not getting it in perfectly formatted.

How is this different from keeping a well-organized Notion database?

A Notion database is organized for human browsing. You navigate to what you are looking for. A compiled wiki is organized for querying. You describe what you need in plain language and the system finds and synthesizes it across everything you have ingested. The difference becomes significant as the knowledge base grows and the connections between things multiply beyond what you can track manually.

Can freelancers query the wiki directly?

That is a judgment call. I give access selectively, usually to longer-term contractors who benefit from the full context. For shorter engagements, I run the query myself and send them the output as a brief. Either way, they get better context than they would have otherwise.

How I Built a Business Brain as a Solo Founder Using Karpathy’s LLM Wiki was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.

Andrej Karpathy posted a thread a few months ago about how he uses LLMs to build personal knowledge bases. Engineers and researchers immediately recognized what he was describing.

Most consultants, founders, and advisors read it and moved on. Nothing in the post connected to running a consulting practice, managing a portfolio of companies, or trying to do serious market research without a dedicated research team. The examples came from academic labs. The tools he named were command-line tools. The entire frame was researcher language.

That is a missed opportunity. What Karpathy built is one of the most practical research systems a business operator could run. The setup is technical, but the underlying system is not. You just need someone to translate it.

This is that translation.

What Karpathy Actually Built

Not a note-taking app. Not Notion with AI features. Not what most people mean when they say second brain.

Here is the plainest version: you collect raw source material into a folder, an LLM reads everything in that folder and compiles it into an organized wiki, and then you query that wiki like a research assistant who has absorbed everything you have ever fed it.

Three layers. Collect. Compile. Query.

The wiki lives as markdown files on your computer, structured into concepts and categories with backlinks connecting related ideas. You never write the wiki yourself. Claude Code builds it, updates it, and maintains it every time you add new material. Your only job is to feed it focused content and ask it the right questions.

What makes this different from every other knowledge system is compounding. The more material you add, the more powerful the query layer becomes. Karpathy’s own wiki on recent research runs at around 100 articles and 400,000 words. At that scale, he asks it complex multi-part questions that would take hours to answer manually.

There is also a feedback loop most people miss. The outputs from your queries get filed back into the wiki. Every time you run a session and get a useful synthesis, you save it back in. Your own research questions and explorations accumulate over time. The wiki gets smarter the more you use it.

The Five Components, Translated for Operators

Data ingest is the input layer. You collect articles, PDFs, reports, and research into a folder called raw. The Obsidian Web Clipper browser extension converts web content into markdown automatically as you read, saving files locally without manual copy-pasting. PDFs and documents go in by drag and drop. You can also set a hotkey to download related images so your LLM can reference them during queries.

The wiki is the compiled layer. Claude Code reads everything in your raw folder and produces organized markdown files. It identifies concepts, writes summary articles for each one, creates backlinks between related ideas, and maintains an index you can navigate. You do not write or edit any of it. The LLM generates and maintains the entire structure.

Q&A is where the return becomes visible. Once your wiki reaches around 80 to 100 articles, you can ask it complex questions in plain language. It reads across the relevant files and returns answers that reading your notes individually would never surface. Karpathy notes this surprised him: he expected to need sophisticated retrieval tools, but the LLM handles navigation well at this scale without them.

Output returns in whatever format you need. Written summaries and analysis come back as markdown files. Presentations come as Marp-format slides. Charts and visualizations come as images. Everything is viewable inside Obsidian, so your entire research environment stays in one place.

Linting is the health check layer. You run periodic passes over your wiki asking Claude Code to find inconsistencies, fill gaps, and surface connections between articles you never explicitly made. This is where the compounding becomes most visible. The wiki finds relationships across your material that you missed reading individually.

How to Set It Up

Three tools: Claude Code, Obsidian, and the Obsidian Web Clipper browser extension.

Claude Code runs the compilation step, reading your raw folder and writing the wiki. Obsidian is the interface where you read the wiki, browse the graph view, and view query outputs. Web Clipper gets web content into your raw folder with a single click while you are already reading.

The full technical walkthrough is in a previous piece. [Link to “How to Build the Knowledge System Andrej Karpathy Uses.”] Get the structure working first, then come back.

Make one decision before you build anything: what domain is this wiki covering? One topic per wiki. Trying to capture everything you read across your entire professional life produces a system useful for nothing. Pick one focus area, build it to 80 articles, and only consider expanding scope once the Q&A layer is returning answers you trust.

What to Feed It, By Role

Most operators make the same mistake when they start. They clip everything because clipping is easy. That is wrong.

Your wiki becomes useful in proportion to how focused its inputs are. The question is not what to collect generally. It is what to collect for this specific wiki, for your specific role.

Consultants should build one wiki per industry vertical, not per client. Feed it sector reports, regulatory filings, competitor earnings calls, M&A news, and the research notes from past engagements in that vertical. When a new client comes in from the same industry, your wiki already carries months of compounded context. You arrive at the first discovery call with preparation that normally takes a week to assemble from scratch.

Founders and startup founders should run two wikis. One for the market, one for the product domain. The market wiki gets competitor product updates, investor thesis documents from relevant funds, fundraising announcements from your space, and market sizing research. The product wiki gets user research transcripts, technical literature, and teardowns from adjacent product categories. Keeping them separate means your queries return focused answers instead of mixing investor intelligence with product architecture questions.

Advisors can organize by sector. Build one wiki per industry your portfolio companies operate in. Board prep that used to take two hours becomes a structured 20-minute session because the contextual depth is already compiled and queryable. You stop relying on what you happen to remember from the last board call.

Small business owners should keep it simple: one wiki for the market, one for operations. The market wiki gets customer reviews from your category and your competitors, local pricing intelligence, and relevant trend coverage. The operations wiki gets supplier relationships, process documentation, and regulatory updates that affect your business. Two narrow wikis consistently outperform one sprawling one.

What to Ask It, By Role

The output quality depends entirely on the question quality. Generic questions produce generic answers. Specific questions with real business context produce answers you can act on the same day.

Consultants get the most from questions like “What are the biggest operational shifts in this client’s sector in the last six months?” and “What was my core argument in the last proposal I wrote for a client in this vertical?” The second question becomes powerful once your wiki holds material from several past engagements. You stop rebuilding your arguments from scratch on every new opportunity.

Founders should lean on questions like “What objections do investors in this category raise consistently?” and “Where are the positioning gaps across my top three competitors?” Feed your wiki competitive material for two to three months and that second question will surface patterns you missed reading the same articles one at a time. The wiki synthesizes across sources in a way your memory cannot.

Advisors should ask “What should I know going into the board meeting for Company X this week?” and “Which of my portfolio companies have exposure to the same market pressures right now?” Beyond five active advisory relationships, the second question is almost impossible to answer accurately from memory alone. Your wiki answers it in seconds, pulling from sector coverage spread across months of input.

Small business owners should start with “What problems in my category has no competitor solved yet?” and “How has my main competitor’s pricing or positioning changed in the last year?” The first question is a product or service development brief. The second is a competitive intelligence summary. Both are assembled from sources you were probably already reading without retaining them in any useful form.

The Step Most People Skip

Run a linting pass every two to three weeks.

Ask Claude Code to scan your wiki for inconsistent or contradictory information, identify topic gaps, and surface connections between articles you never explicitly linked. This is where the compounding becomes concrete rather than theoretical. The wiki finds relationships across your material that you missed when reading each piece individually.

Karpathy uses this step to identify new questions worth investigating and connections between research threads. For a founder, a linting pass might surface two competitor moves that point at the same market shift. For a consultant, it might link a regulatory filing from three months ago to a question a client raised last week. Over time, the linting passes also reveal where your reading has blind spots relative to your domain.

Where This System Falls Short

Comfort with Claude Code is a hard requirement. If you have never used a command line tool, the setup will take longer than an hour and there will be moments where nothing works and there is no support chat to fall back on. That stops a lot of business operators before they get any value.

The wiki is only as good as what you feed it. Clipping low-quality content produces low-quality answers. No configuration compensates for weak inputs. The discipline is in the curation, and curation requires judgment the tool cannot substitute for.

There is no collaboration layer. This is a solo research engine. Sharing outputs with a colleague means exporting files manually, which works but is not seamless.

The useful threshold also takes time to reach. At 20 articles, your wiki is a better-organized folder. At 80 to 100 articles, it starts functioning like a research assistant. Plan for six to eight weeks of consistent input before the Q&A layer returns answers you trust enough to act on.

If You Want a Version That Is Already Built

When you spend all day reading stuff like this, trying to remember it later. Constella lets you instantly clip sections and entire articles as you go.

This then all gets added to your self-organizing AI brain.

Now, when reading future articles, or writing, you can recall related ideas.

Use it to find the right clippings and thoughts you need when writing.

Or for inspiration when you go “hm, I see a connection here to something I read before.”

What we built at Constella is for operators who want the output layer without constructing the infrastructure. If the Claude Code setup in this article felt like a hard requirement you are not ready to meet yet, this is the alternative.

The research canvas maps your sources as nodes in a visual graph, extracts implications automatically, and surfaces research gaps across everything you load into it. The compounding effect, without the command line setup or the six-week ramp.

Karpathy’s LLM Wiki System, Explained for People Who Run Businesses. And How To Use Them. was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.

Richard Branson carries a notebook everywhere he goes. Bill Gates spends two weeks every year doing nothing but reading, writing, and thinking. Jeff Bezos built an entire written communication culture at Amazon that shaped how the company made decisions for decades.

Note-taking is one of the most common habits among successful founders. It’s documented, studied, and almost universally recommended. The founders who do it consistently build on their own thinking over time. They catch patterns others miss and hold onto what most people lose in the noise of running a company.

But most founder notes are useless.

And it lies on what they’re actually capturing.

What Most Founders Get Wrong

The typical founder’s notes look something like this. Meeting summaries. Key decisions. Quotes from books. Ideas they don’t want to forget. Lessons from mistakes they want to avoid repeating.

All of it is information. That’s the problem.

Information is the least valuable thing you can put in a note. You already remember most of it, or you can find it again in thirty seconds. What you can’t find again is what you were actually thinking at the time. What you were connecting it to. Why it mattered to you in that specific moment, at that specific stage of your company.

An idea without context is almost worthless when you come back to it months later.

Say you wrote “build the team before you scale.” Good advice. But why did you write it? Was it after you burned yourself out trying to do everything alone? Was it after watching a competitor collapse because their founder couldn’t let go? Was it from a conversation with an investor who had seen it happen dozens of times?

The reasoning is what makes the lesson stick. The context is what makes the idea useful to you, in your situation, for your company.

The same goes for mistakes. You can write “don’t do that again” all you want. Without understanding why, you’ll probably do it again. Behavior doesn’t change based on information. It changes based on understanding.

How to Pull More Out of Your Head: Freewriting

The instinct when taking notes is to extract only what matters. The clean summary. The key takeaway. The headline version.

But the real insight is usually buried underneath. You don’t reach it by summarizing. You reach it by writing through it.

Freewriting is the practice of writing without filtering as you go. It isn’t traditional journaling. It isn’t structured brainstorming. It’s closer to thinking out loud on paper, following a thread wherever it goes without stopping to clean it up along the way.

The goal isn’t a polished note. The goal is to arrive somewhere true.

Here’s how it looks in practice. Say you just finished a chapter in a book on hiring. Instead of copying out the key points, close the book and write about what you actually think. How does this connect to what you’re dealing with right now in your company? Where do you disagree? What does it remind you of from your own experience? Let it run without stopping to decide whether it’s useful.

The same applies to the day-to-day of running a startup. When something significant happens in your company, don’t just log what happened. Whether it’s a milestone, a hard mistake, or a conversation that changed how you see a problem, write about it. What does it mean? What were you thinking when it happened? What would you do differently?

Five minutes of freewriting after a hard day will teach you more about yourself as a founder than a year of bullet-pointed summaries.

Like when you spend all day reading stuff like this, trying to remember it later. Constella lets you instantly clip sections and entire articles as you go.

This then all gets added to your self-organizing AI brain.

Now, when reading future articles, or writing, you can recall related ideas.

Use it to find the right clippings and thoughts you need when writing.

Or for inspiration when you go “hm, I see a connection here to something I read before.”

How to Make Your Notes Actually Useful

Notes are only valuable if you come back to them. But even when you do read them back, a note read in isolation rarely gives you much. The real value comes from connecting a current thought to an older one. Seeing how your thinking has changed. Noticing that the problem you’re dealing with today is the same one you were working through a year ago, just at a different scale.

That connection is where the real insight lives.

When I’m working through something, I don’t scroll through old folders looking for relevant notes. I describe what I’m thinking and let the system surface everything related onto a single canvas. Past reflections, older notes, ideas I had half-forgotten. I can see how a decision I’m making today connects to something I wrote eighteen months ago, or how a thought I had while reading maps directly to what’s in front of me now.

This is the difference between storing your thinking and actually using it. Most systems treat your notes as a library. What actually helps is a thinking environment, somewhere you can see how your ideas connect, not just where they’re filed.

The payoff isn’t just better memory. It’s better understanding. The goal of keeping notes as a founder isn’t to remember more. It’s to think more clearly. To make decisions with the full weight of your own experience behind them, not just what happens to be in your head right now.

What Good Notes Actually Do

The founders who get the most out of note-taking aren’t the ones with the most organized system. They’re the ones who use their notes to understand themselves better.

Write your thoughts, not just your ideas. Write your reasoning, not just your conclusions. Build a way to connect what you’re learning now to what you’ve figured out before.

Your notes should make you a sharper founder. If they’re not doing that, you’re keeping records, not building understanding.

Most Founders Take Notes The Wrong Way. Here’s How to Do It Right. was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.

Ever since I used an AI second brain, meeting prep that used to take me an hour now takes five minutes. Market research that used to eat half a day takes twenty minutes. The question about past decisions that used to derail an afternoon now resolves in one query.

When you’re running a business, most of the work of running is looking for stuff and context that are scattered across different apps and piles of chats.

It could be the decision you made about this exact problem in the previous month, and the reasoning behind it. Or he article you read in February that’s suddenly relevant to the strategic call you’re making today.

You know it’s all in there somewhere, buried in Slack threads, scattered across Notion pages, or sitting in old email chains. For some reasons, you just can’t get to it when it matters.

This is what turns most founders into the bottleneck in their own business, even when their strategy is sound.

This is the reason why I decided to build an AI second brain. It’s structural. My business now has access to everything I know.

This piece is about how I actually use and build it.

How I Use It

How 1: My Team Asks My AI Second Brain Before They Ask Me

I connected my AI second brain to a simple chat interface my team can use. Instead of sending me a message about everything, they can query it for something I’ve documented and every decision I’ve already explained.

This way, most things are answered without me being in the loop.

This sounds like a small change. But it isn’t.

Before this, I was the answer machine for my employees. Every question with a known answer came back to me and every piece of context that lived in my head got pulled out by someone who needed it.

I was losing hours every week to questions that only required me to find the context and send it back. That’s why it was the first thing that made sense to automate.

Now my team gets their answers directly from the system. I spend my time on the questions that genuinely require fresh thinking.

How 2: I Make Decisions With Full Context

It’s not what most people do, even people with sophisticated note-taking systems. They store things. They don’t retrieve them before they decide.

The thing with conventional note-taking system is while they store a lot of things, you mostly couldn’t retrieve what you’ve saved unless you know the keywords. Retrieval depends on memory.

With my AI second brain, however, I can query it for everything relevant before making any decisions. It could be past conversations on the topic, or the version of this problem I solved last year and what I learned from it.

When I make a pricing call, I’m pulling from every pricing conversation I’ve had in the last three years and every market signal I’ve noted. This allowed me to make decisions from a position of real context.

How 3: I Never Lose the Thread on a Client or Conversation

Before any client meeting, I pull the thread. The commitments I’ve made and the context shifts that have happened over the relationship. It takes about thirty seconds.

You build context with clients over months. Then a new conversation comes up, and you’re scrambling to remember what was said in the kickoff call, what they promised, what you committed to. The relationship suffers because your memory is the bottleneck, even though the information existed somewhere. This is exactly why the best client management tool isn’t a CRM.

I used to walk into meetings half-prepared, hoping the right details would come back to me. They sometimes did. They often didn’t. The relationship paid the cost either way, because the client could feel when I was reconstructing things in real time.

Now the business doesn’t depend on my recall. I show up sharper because the recall is no longer the work. The work is the thinking, which is what I’m supposed to be doing anyway.

How 4: I Turn What I Read Into Working Knowledge

I read a lot. Most professionals at my stage do. The honest truth is that almost none of it used to stick. I’d finish a great book, highlight things, and within a month I couldn’t tell you the core argument. There’s a real problem with being well-read if none of the reading actually compounds into how you think and decide.

My AI second brain captures everything I read. It connects new ideas to old ones. And when something I read becomes relevant to a current problem, it surfaces. I don’t have to remember where I read it. I don’t have to remember that I read it.

This is the part that feels closest to magic, even though it isn’t magical at all. The connections I never had the time to make explicitly get made for me.

How 5: I Stopped Solving the Same Problems Twice

Every business has recurring problems. Team friction. Pricing pressure that you have to keep negotiating.

For years, I solved each one fresh. Same problem, same hours of thinking each time. I never got better at solving these problems either, because I was building no real foundation. Each round started over. This is also [how expertise narrows thinking] instead of widening it. Without a system that builds on what you’ve already worked out, expertise turns into the same handful of moves repeated.

Now when a familiar problem shows up, I start from what I learned the last time. The thinking compounds. The business stops being a place where the same lessons have to be re-learned every year.

How to Make the AI Do the Heavy Lifting

This is where the AI second brain breaks from the older generation of tools.

Tools like Notion and Obsidian are archives. You retrieve what you explicitly stored, if you can remember the right page or the right tag. The system works as long as you maintain it. Most people stop maintaining it after about six months.

An AI second brain works differently. You feed it information. It synthesizes across everything you’ve put in. You ask questions in plain language. You don’t manage tags or folders. This is essentially the knowledge system Karpathy uses, applied to a business instead of a research project.

You spend all day reading stuff like this, trying to remember it later. Constella lets you instantly clip sections and entire articles as you go.\

This then all gets added to your self-organizing AI brain.

Now, when reading future articles, or writing, you can recall related ideas.

Use it to find the right clippings and thoughts you need when writing.

Or for inspiration when you go “hm, I see a connection here to something I read before.”

The bigger shift is that it finds connections you never made yourself. The note you scribbled from a client call last March turns out to be relevant to the strategic question you’re working on today. You didn’t link those two things. The AI did.

The most useful version of this for me has been working on a canvas where every source becomes a node. New papers and articles get pulled in automatically. The system surfaces what’s relevant without me having to ask the specific question. I see the shape of my own thinking, and I see where it has gaps. This is also how I do market research now, which used to be a multi-day slog.

The more you feed it, the more useful it gets. That’s the part conventional tools never delivered. They got heavier the more you added. This gets sharper.

Most people running businesses I respect are still trying to do it from memory. They have notebooks. They have apps. But they’re still bottlenecked by their own recall.

Your business deserves access to everything you’ve learned. But the bottleneck was always human memory.

The AI second brain is how you fix that.

How I Use My AI Second Brain To Run My Business was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.

TL;DR

• Knowledge management spent a decade solving a storage problem when the real problem was synthesis.
• AI has shifted the category from filing cabinets to thinking partners.
• The tools winning in 2026 surface and connect knowledge instead of just holding it.
• Organizations that understand this shift are quietly building a compounding advantage.

The latest knowledge management AI news in 2026 marks a clean break from what the category used to be. Software in this space is no longer organized around capturing information. It is organized around making information move, connect, and surface at the moment someone needs it. That shift changes what a knowledge tool is for.

We built enormous libraries and called it knowledge

For most of the last decade, knowledge management was a filing problem the industry kept naming as a thinking one. Notion, Confluence, SharePoint, internal wikis, every team chat thread you ever pinned. They promised that capturing more would lead to knowing more, but the result was a graveyard of notes.

Walk into any company that has been around for more than five years and ask where their knowledge actually lives. Some of it is in a wiki nobody reads, or in Slack threads that scroll past faster than they can be processed.

The architecture made sense in 2015. Computers were getting cheaper, search was getting better, and the dream was that if you could just persuade people to write things down, the rest would follow. But writing things down was never the bottleneck. The bottleneck is connecting what was written down to whoever needs it on a Tuesday morning at ten.

Companies have been keeping commonplace books in digital form for years now without ever fully understanding what such a practice was actually meant to do.

The form was preserved. The function got lost.

What AI actually changed about knowledge management

Knowledge management news in 2026 has been dominated by chatbot announcements, and that is part of why the bigger shift is being missed. The companies covering AI loudest are still focused on chat interfaces. The interesting work is happening one layer down.

AI did not show up in this category as a chatbot bolted onto an existing wiki. The tools that are winning treat AI as the substrate. Knowledge gets ingested, parsed, embedded, connected, and held in a graph that updates every time someone writes something new. A search is no longer a string match, but a synthesis that reads across documents, chats, emails, and notes the user has long since forgotten about.

This is a different category. The old generation of tools held information. But the new generation holds context.

The deeper structural point is the part most coverage misses. AI has become the foundation layer of these products instead of a feature added on top. Products built before this shift are trying to bolt on what the new ones were architected around. That gap is wider than it looks from the outside, and it is the same gap YC has been quietly orienting its 2026 cohort toward. Foundations don’t get added later. You build on them or you don’t.

The companies that grasp this are rebuilding the category. The promise that AI was going to amplify human thinking is starting to land in places where the work actually happens.

Where knowledge management tools are heading in 2026

The latest knowledge management trends share a common direction. Each of them is a step away from the filing cabinet model, and a step toward something closer to organizational memory that thinks for itself.

1. The death of folder structures

The first trend is the quiet death of folder structures. Knowledge graphs are replacing taxonomies as the way information is organized. People stop creating folders and tags because the system maintains its own map of how ideas connect.

The user writes. The system structures.

2. Proactive surfacing

The second trend is proactive surfacing. Knowledge tools used to wait for you to search. The new generation? It pushes relevant context to you before you ask, in the document you’re writing, the email you’re replying to, the meeting you’re walking into.

The interface for knowledge moves toward where the work already lives.

3. Integration over isolation

The third trend is integration over isolation. Knowledge management software news in 2026 is dominated by tools that hook into Slack, Gmail, Notion, Obsidian, calendar, code repositories. The tool nobody opened was the wiki.

The tool people use is the one that meets them inside the apps they already have.

4. Organizational memory is becoming its own product category

A fourth direction worth naming is organizational memory becoming its own product category. Onboarding, expertise capture, decision archives, project history. These used to be operations problems with no good software. They are starting to have it.

If you have been watching this space, you can see the categories of personal knowledge tools and enterprise knowledge tools beginning to converge. What used to live as a personal productivity question is becoming an institutional one.

The organizations that get this are pulling ahead

The compounding consequence of the shift is the part that gets underappreciated.

Every company has the same hidden tax. A senior engineer leaves and takes context with them. A product manager moves teams and the institutional memory of three failed launches goes with her. A consultant ends an engagement and the client retains a deck instead of the thinking that produced it. The cost of that knowledge loss is invisible until it is staggering.

Companies using new-generation knowledge tools are starting to retain that knowledge in a way the old tools never could. They are doing it by structuring what gets written so the context survives the person. Writing more down has been the failed strategy of the last fifteen years. Structuring it as it gets written is a different proposition entirely. The senior engineer’s reasoning, not just her code commits. The product manager’s edge cases, not just her launch deck.

The competitive picture this creates is uneven and getting more so. Two companies in the same industry, with the same headcount, with the same revenue, can have wildly different rates of organizational learning depending on whether their knowledge compounds or evaporates. Experience matters less when it cannot be transferred, and the firms that have figured out how to transfer it are accelerating away from the ones that have not.

This is institutional intelligence dressed up as a productivity story.

Why I’m building in this space

I have spent the last few years working on a knowledge management product called Constella because the version of this category I want to use does not yet exist at the level I think it should.

You spend all day reading stuff like this, trying to remember it later. Constella lets you instantly clip sections and entire articles as you go.

This then all gets added to your self-organizing AI brain.

Now, when reading future articles, or writing, you can recall related ideas.

Use it to find the right clippings and thoughts you need when writing.

Or for inspiration when you go “hm, I see a connection here to something I read before.”

The problem I keep coming back to is the connection problem. Most of what we know is held in fragments scattered across different surfaces. Papers we read but did not annotate. Conversations we had but did not write down. Documents we drafted and then archived. Notes we made on a phone screen and then lost in the scroll. The fragments are not the issue. The issue is that nothing knows about all of them at once.

The bet underneath the product is that the most valuable knowledge tool of the next decade is the one that holds an entire researcher’s or founder’s body of thinking as a single connected graph and treats every new question as an excuse to traverse it.

The category is moving in this direction with or without us. I would rather be inside the work than watching it.

Knowledge management has outgrown note-taking was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.

Ask any founder with ADHD and they’ll say it’s their superpower. I get it. As a founder with diagnosed ADHD myself, I said the same for years.

But here’s the thing.

It works for a while. Hyperfocus ships the demo. The restlessness keeps you moving when most people would have stopped. You raise on the energy of it.

Then the company grows. Payroll, contracts, customer tickets piling up. The same ADHD brain that built the product can’t run the growing operation.

Took me a few years to figure out how to actually make ADHD pay off as a founder. Once I did, things got a lot easier.

This piece is about what I started doing instead.

The Real Cost Shows Up When You Can’t Be in Just One Place

For years I treated ADHD like fuel, something to rely on when I needed to push through.

But that was enough when I was a solo founder.

Then we had employees, customers, contracts, a board to answer to. Hyperfocus only helps with the one thing you decide to focus on. Everything else has to wait.

I’d give a day to one product feature and look up to find an inbox I hadn’t touched, a contract still on my desk, customer emails three days deep. Every time I produced one good thing, a few other things were quietly going wrong while I was working.

My ADHD started being a liability rather than a superpower.

And it showed up the moment your attention has to be in more than one place at once.

What I Changed

So here’s what I changed.

Every morning, before I opened anything, I picked one thing. The sharpest hour of my day went to that one thing, no exceptions.

The reason it mattered is that ADHD chases urgency. The brain wants the new message, the open thread, the thing someone just Slacked you about. Following that feels like work. It almost never is.

I also stopped telling myself I’d remember everything. I’d keep it all in my head. I was the founder, I was supposed to hold it all at once.

But I couldn’t.

ADHD just makes that more obvious sooner than it would for anyone else.

So I built a simple system around myself.

• A weekly review where I write down everything that needs to happen the next week.
• The morning pick for the one thing.
• A capture habit where anything that pulls at my attention goes into a list instead of pulling me off track right now.

It might not look like a brilliant system. But it holds the things I couldn’t, and my brain is free to go deep without losing the rest of the company.

Stop Trying to Be the Brain for Everything

For a long time I thought being a good founder meant holding the whole company in my head. If it left my head, it would be lost.

The problem is that an ADHD brain is the worst possible place to keep a company.

Things did get lost — customer emails, decisions and promises I made on calls, I’d remember them at two in the morning and feel sick about it.

So what worked was getting honest about what my brain was actually for. It’s good at chewing on hard problems, going deep on something new. But it is bad at remembering, tracking, and following through.

So I stopped trying to make it do those jobs.

Anything that needed remembering went into a system. Anything that needed following through went to someone on the team whose brain works the way mine doesn’t. My job became thinking and deciding. Their job became making sure the decisions actually happened.

It sounds obvious written down. It took me years to stop fighting it.

Treat Your Energy Like the Real Budget

Time isn’t the constraint for an ADHD founder. The actual constraint however, is the energy.

I have maybe four hours a day where my brain is actually sharp. If I burn those hours on something that doesn’t need them, the rest of the day is just maintenance.

So I stopped scheduling around time and started scheduling around energy. Sharp hours go to the work only I can do. Meetings go to when my brain is already coasting. Admin gets saved for when I’m tired, because tired me can still answer emails. Tired me cannot design a product.

Because the most difficult part is protecting those hours even when other people want them. I might get pushback from people who want morning meetings, and they get my afternoon instead. But the company runs better for it.

The Quiet Part

A lot of the founders winning quietly with ADHD are also medicated. Nobody says this out loud much.

I’m not going to tell anyone what to do with their body. But the story that you can run a growing company on willpower and hyperfocus alone is mostly told by founders who haven’t tried to do it yet.

For me, medication didn’t change what I’m capable of. It changed how often I have access to it. The sharp hours became more reliable. I could choose what to focus on instead of letting whatever was loudest in the room win.

If you’ve never talked to a doctor about it, that conversation is worth having.

ADHD didn’t become useful when I started calling it a superpower. It got closer to leverage when I stopped.

The trait helps with parts of building a company. The cost is also real, and it shows up later than the benefit does. The work is building the life around it that lets it actually pay off.

How I Learned to Make Neurodivergence Pay Off as a Founder was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.

YC just released their Request for Startups for the Summer 2026 batch yesterday.

The first thing they say on the press release is that AI as stopped being a feature and started being the foundation.

That strongly indicate what type of startups they want to fund. And it makes sense, it’s where the AI industry is heading, and also where the money is going.

But here’s the thing. Most products that call themselves AI-native are still using AI as a feature. They think they’re building AI as a foundation, but they don’t.

So what does AI being a foundation actually means?

What People Get Wrong

Ask most founders what makes their product AI-native and they'll talk about architecture. They're running on LLM,fine-tuned a model on proprietary data, or built a retrieval pipeline. These matter matter, for sure. But they're not what makes something a foundation.

I've seen products with impressive infrastructure that are still, functionally, just a better search bar. The AI is doing something sophisticated in the background, but the user is still doing the same things they were doing before. Well, just a little bit faster.

But that's technically still a feature.

The thing is, what makes AI a foundation isn’t about the technical. Instead, it’s behavioral.

It’s about how users actually move through the product — what they do, and what they no longer have to do, or what becomes strange about the way they used to work.

In other words, when AI is the foundation, the workflow itself stops making sense the old way. The product doesn’t improve a behavior, it replaces the need for it. But when AI is a feature, the user’s workflow stays the same and AI only helps with a step.

One example is Notion. They have AI built into their workspace. They can create pages, tables, and summaries for you. But if you remove the AI, does the product still exist? The answer would be yes. That said, AI is still a feature.

And that’s what separate whether AI is a foundation or just a layer on top of something that already existed.

Most founders are building the layer and calling it the foundation. That's what YC is pushing back on.

What This Means If You're Building Now

The question most founders ask is where AI fits into their roadmap. But that's the wrong question. It assumes the product already exists and AI is something you add to it.

The right question is what does this category look like if you rebuild it from scratch, with AI as the assumption from day one?

To answer that, you’d have to start with a blank page and asks what becomes possible, what becomes unnecessary, what the user actually needs to do once AI is doing the rest.

The founders YC seems to be betting on, based on the RFS, are the ones who actually sat with that question long enough to throw out the assumptions of their category.

That's a much harder thing to do. It means accepting that some of what you thought was essential to your product was only essential because the old tools made it essential.

How I'm Building It

I’m in the middle of this shift myself.

The startup I’ve been building, Constella, used to be a PKM tool. A place to capture notes, link them, organize them. AI was in there, but it was a feature. It helped you summarize, search, suggest tags. Useful, but the shape of the product was still the same shape every PKM tool has had for the last fifteen years. The user was still doing the heavy lifting.

That’s not what we’re building anymore.

The category we’re building in now isn’t note-taking. It’s a knowledge system. And the difference matters, because a knowledge system isn’t something you organize. It’s something you talk to. Something that pulls from what you’ve captured, what you’ve read, what you’ve been working on, and connects it without you having to ask twice. The user’s job is to feed it and to think with it. Everything in between is the system’s job.

That reframing is what AI as a foundation actually looks like in practice.

Take research. The old workflow looks the same whether you’re a PhD student, a founder doing market research, or an analyst writing a report. You collect sources. You take notes. You organize the notes. Later, when you need to write or decide something, you go back, search through your notes, retrieve the right ones, and try to remember how they connect.

Five or six steps, all manual, all dependent on your memory and your filing system holding up months later. It rarely does.

Now ask the foundation question. What does research look like if AI is the assumption, not the addition?

The answer is no longer a notes app with a chat panel. It is a canvas where every source you bring in becomes a node in a graph. Where the connections between sources surface automatically because the system has actually read them. Where you can ask a question across everything you’ve ever captured and get a synthesis that pulls from the right places, with the original sources still there to verify. Where the system tells you what’s missing, where the gaps in your research are, what you should read next.

That’s the product I’m building. And it started from asking what a knowledge system should look like when AI is doing the work of organizing, retrieving, and connecting.

The implications go further than research.

Once the system can read what you’ve captured and understand it, the integrations stop being about syncing files. Constella connects to Obsidian, Zotero, Readwise, Notion, Slack, Gmail. Not to import data into another silo, but because the knowledge system needs to see across all of them. Your reading lives in one tool. Your writing in another. Your conversations in a third. A real knowledge system has to span all of it, because that’s where your actual thinking happens.

Five years ago, I wouldn’t be able to build that product. And that’s simply because the foundation didn’t exist yet. It’s AI that makes the entire shape of this product possible. If you take it out, there’s nothing left.

Because AI is the foundation and everything else is built around it.

None of this is easy to get right. I’m still figuring it out. The shift from feature to foundation isn’t a one-time decision you make on a whiteboard, it’s something you keep running into every time you build the next thing.

But I believe it’s the work that’s actually worth doing right now.

AI Is A Foundation, Not A Feature. That’s What YC’s Summer 2026 Startup Request Is Really About. was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.

Managing multiple projects at the same time has never been the hard part for me. What I found difficult is that every client runs on a different set of tools. One lives in Slack. Another sends everything through email. A third has their own project portal they built in Notion.

And I’m expected to know what’s happening across all of them, all the time, without missing anything. Gosh.

The constant switching between apps to check progress, grab context, or find something a client mentioned two weeks ago took up more of my working hours than I expected.

And the worst part was when small details got missed, decisions or agreements made in a chat thread got buried under fifty other messages.

Then the misunderstandings happened.

And every time one did, I’d spend twenty minutes tracing back through four different apps trying to figure out where things went wrong.

That gave me more headaches than anything else in the job. So I tried to fix it.

I’ve Tried Just About Everything

A Personal Notion Doc

Of course I started with Notion. It’s literally one of the best tools out there to manage your information. So what I did was that I created a page for each client, wrote down anything important after every call or message, kept it organized by project.

Weeks went by I felt great. I’d go into client calls knowing things I would have otherwise forgotten. A preference someone mentioned in passing. Or a concern they raised once and never followed up on. I felt on top of things in a way I hadn’t in a while.

But keeping it updated required a kind of consistency that real workloads don’t allow for. A few busy days in a row and I’d fall behind. I’d tell myself I’d catch up later.

But lter never came. A month in, half the entries were outdated and the other half were so sparse they weren’t useful. I ended up going back to the original apps as it was faster than trusting what I’d written down.

A Master Tracker

So I thought the problem was structure. That if I built something more organized, with a clearer system behind it, I’d actually stick to it.

I spent a whole afternoon building a master tracker. Every client had their own section. Project status, key contacts, important decisions, preferences, outstanding items. I color coded it. Kept everything properly formatted.

For a while it genuinely worked. Starting my mornings by scanning the tracker and getting a clear picture of all my clients at once brought a calm I hadn’t felt in a long time. I started to think I’d finally cracked it.

Then a busy stretch hit and updating the tracker became the first thing that got skipped. And it makes sense. Because there was always something more urgent. Then the tracker quietly became a record of a calmer time, not the reality I was actually living in.

Alright, It’s Time For The Actual Project Management Software: Asana

After two failed attempts at building my own system, I figured the answer was to use tools that were actually built for this — the actual project management software. I set up workspaces for each client in Asana, built out task lists, logged updates, tracked deadlines.

But tasks were never the problem. Context was, Asana can tell me a deliverable is due Friday. It cannot tell me that a client mentioned three weeks ago in a Slack message that their leadership team just changed and everything might be up for review. That kind of detail doesn’t live in a task. It lives in a conversation that happened once, in an app I haven’t opened since, and it matters enormously when I’m walking into a call.

And all the other attempts

I tried everything else too. Setting reminders. Starring important messages in Slack and Gmail so I could find them later. Asking clients if they’d consider consolidating into one tool. Writing quick end-of-day summaries so nothing would get lost overnight. Each of these helped a little on the days I actually did them.

But they all had the same problem.

They required extra effort on top of an already full day, and the moment life got busy, they were the first things to go.

The Actual Problem I Didn’t Realize

At some points I was really frustrated and thought that probably I just had to keep it the conventional way — deal with the fact that I’m working through a lot different apps at once.

But then I started asking question, what was the actual problem? Why did the fix collapse at some points and especially when things got overwhelmed.

The answer I came up with was that every fix I tried assumed the problem was organization. That if I just built a cleaner system, maintained it more consistently, or chose the right tool, the details would stop slipping.

But organization was never the issue.

The real issue was fragmentation. The information existed, it was always there, sitting in some Slack workspace, some email thread, some Notion page a client shared access to months ago. But they all lived across too many places at once, and there was no way to reach all of it without switching between apps and losing the thread every time I did.

Adding another tool just gave me one more place to check.

Finally, The Fix

When you’ve worked with enough projects, you just started accepting the facts that clients aren’t going to abandon their tools, and asking them is a losing battle.

What needed to change was how I could get to it. One surface that could reach into all those different places and pull out what was relevant, when I needed it, without me having to go hunting for it manually.

I couldn’t find anything that did exactly that, so I built it myself. That’s how Constella came about. It connects to the tools most clients already use — Notion, Slack, Gmail, Google Drive — and when I need context on a specific client or project, I just need to ask. It pulls from wherever that information lives. I don’t have to remember which app it was in or scroll through weeks of chat history.

It’s another different app, yes, but it surfaces what’s relevant and I can get on with the work.

There’s no new system to maintain, no manual copy-pasting, clients keep using their tools, and I finally stopped losing hours.

What This Actually Changed

Walking into a client call now feels different. The context is accessible when I need it instead of buried somewhere I’ll only remember to check after the call ends.

The fragmentation was always the problem. And no amount of better habits was ever going to fix something that was structural from the start.

How I Finally Fixed My App Switching Problem Managing Multiple Clients was originally published in The AI Second Brain on Medium, where people are continuing the conversation by highlighting and responding to this story.