Why Most People Set Up Their Bots Wrong
Imagine the following. A company hires someone, takes them through an orientation, and hands them a business card that says “Marketing Manager.” On day one the new hire sits at their desk, card in hand, and wonder: Managing what? For whom? In what tone? And what should I never, ever do?
A title is a start. It isn’t a job.
The same is true for AI bots. Lately I’ve been working with a small team of bots, each handling a very focused set of responsibilities and roles. The more I use them, the more I admire how they come to know their jobs. It isn’t clever. It’s simple. There are three layers, each does one thing, and each builds on the last.
It starts with the bot’s name.
The Sign on the Door (The Name)
Name a bot “The Marketer” and it already has a rough idea of its role. It knows it deals with growth, promotion, and audiences. Pretty neat stuff, one word of intent goes a long way.
But a name is only a label. “The Marketer” doesn’t know your brand voice — whether you prefer a bold, disruptive tone or a polished, corporate one. It doesn’t know which personas you are targeting, which platforms you prioritize, or that every campaign must include a specific call-to-action. It knows the job title. It doesn’t know the job.
Useful, necessary, and not remotely enough.
The North Star (The Description)
The description is where most of the real role lives. It’s the bot’s purpose, persona and scope, written in plain language: who it works for, what it’s for, how it should sound, and (just as important) what it should hand off to another bot instead of trying to do itself.
If the name is the sign on the door, the description is the job description pinned to the wall. If you’ve read my earlier pieces on .md files and PRDs, you’ll recognize the idea: write down what good looks like before anyone starts working.
Here’s the part I find truly beautiful. The description isn’t code. It’s prose. Anyone who can write a clear brief for a human marketing agency can write one for a bot. No “prompt engineering” wizardry required.
The Notebook (Memory)
The description tells a bot who it is. But what about everything you learn together along the way?
That’s memory: the notes a bot keeps that persist across conversations, even after a chat is cleared. Tell a bot something lasting (“never use the word ‘synergy’ in our copy,” “our primary target is mid-level product manager”) and it saves it as a fact and applies it from then on.
No more repeating your brand guidelines every Monday morning. No more re-explaining your target audience to a tool that should already know them. The bot doesn’t just follow its description. It learns your preferences on top of it.
As I describe in my article Why Your Agents and Bots Forget the Small Stuff, a lean, living set of rules beats an ever-growing pile of context. Memory is where that set of rules lives.
Mine Versus Ours (Two Kinds of Memory)
Here’s where the structure gets really elegant. There are two kinds of memory:
Get Alex Yampolsky’s stories in your inbox
Join Medium for free to get updates from this writer.
A bot’s own memory is just for that bot. For example, The Marketer keeps notes on the preferred ad hooks and the campaigns that performed best last quarter. The SEO bot doesn’t need them, so they stay with The Marketer.
Shared memory is seen by all of my bots: my name, my time zone, my do-not-contact list. Facts about me, not about any one job.
Why does that matter? Because one rule given to one bot carries over to the whole team. Tell one of them “we are rebranding our primary color to deep navy,” and every bot knows it. Not because you told each one, but because the rule is about the business, and it lives where all of them can read it.
It’s the difference between telling one employee and updating the company handbook.
Why It Works (The Glass Box)
Let’s pause for the inevitable doubts. Isn’t this oversimplified? Shouldn’t something this powerful be complicated? Don’t you need a specialist to set it up?
No, no and no.
Every layer is readable. You can open a description and see exactly what a bot was told. You can see what it remembers and correct it. Nothing hides behind a curtain. It’s a Glass Box, not a Black Box, and that transparency is the whole point. When a bot suggests a campaign that feels “off,” you don’t guess why. You check the label, the brief and the notes, and fix the one that’s off.
That’s what highly logical design looks like. Each layer has one job:
- The name says what the bot is.
- The description says how it should do the work.
- Memory says what it has learned about you.
The bot helping you strategize your growth is the proof. The Marketer works well not because of its name, but because it was set up with your brand persona, your target audience, and a history of what has actually converted, and then kept learning from there.
The Bottom Line
We tend to assume good AI has to be complicated. Sometimes the best structure is the one a beginner can understand in a minute.
It starts with the name. The description gives it purpose. Memory makes it yours over time.
The name is the label. The description and memory make the bot good at its job.
So give your bots a name. Then give them a reason. Then let them get to know you.
To learn more visit www.AlexYampolsky.com
