Context is the whole thing

There's a reason the about page on this site says that. It's not a tagline. It's the thing I wish someone had told me before I spent months wondering why AI tools felt inconsistent.

The prompt matters. The tool matters. But context — what the AI actually knows about you, your situation, your goals, how you think, how you communicate — is what determines whether you get a generic useful answer or something that actually fits. Same prompt, different context, different output. Every time.

Here's what took me a while to understand: context isn't one thing. It's several things, and they have different jobs.

The clearest illustration I have of this: I was taking an AI prompting course that assigned generic, neutral prompts — designed for anyone. One exercise was about a manager explaining an RTO mandate to their team. I copied the prompt and pasted it into my regular Claude chat, forgetting that Claude had months of context about me at that point. The response I got back was not the exercise answer. It was Claude telling me it was going to flag what it was noticing first — that the prompt read like a case study, that it didn't sound like something I was actually navigating, and that if any of it was real, the answer would be different. Direct quote:

My AI was side-eyeing me from across the internet. I had to open an incognito window — a fresh session with no prior context — just to complete a homework assignment without getting called out. That's when it landed: I hadn't just been using a tool. I'd been building one. Slowly, conversation by conversation, without fully realizing it.

That's the thing about context. Once it's there, you feel it most clearly when it's suddenly gone.

Most tools have some version of these four layers. They're not the same thing and they don't do the same job.

Here's the part that surprised me: I didn't arrive at my context documents by sitting down and thinking hard about how I work. Most of it came out through conversation.

Some things I already knew about myself. Some came from excavation — therapy, reflection, the kind of thinking you do when you're forced to slow down. And some just surfaced. A conversation about why a prompt wasn't working turned into a realization about how I process information. A session on something completely unrelated ended with an insight I didn't have going in.

The AI didn't give me those insights. But the conversation created the conditions for them to surface. That's a different thing, and it matters.

What I'd tell someone starting out: don't try to write the perfect context document from scratch. Have the conversations first. Let the material come up. Then capture it. You'll know more about how you work after four weeks of real use than you could have articulated before you started.

This is the part people resist, because it feels inefficient: the conversation is the work.

Yes, it costs something to arrive at a useful output through dialogue rather than a fully-formed prompt that leaps from your head fully formed. That cost is real. But what you get back is more than the output — you get the thinking that produced it, the refinements that made it better, and often something you didn't know you needed until it showed up.

A prompt written in isolation is a guess about what you need. A conversation lets you find out.

This doesn't mean you never write a prompt. It means you earn your prompts. The best ones come out of conversations, not the other way around. Your mileage may vary — but I haven't found a shortcut that gets me somewhere better than the long way does.

You will not be the same person in four weeks. Your situation changes, your priorities shift, your understanding of what you're doing deepens. Context that was accurate in January can be quietly wrong by March — and quietly wrong is worse than obviously wrong, because the tool will use it confidently.

Make updating your context a habit, not an afterthought. When something changes — a job, a project, a significant realization about how you work — update the relevant layer. Memory, instructions, context file, whatever holds that information. Keep it current.

The most important thing about your context is that it keeps being true.

If you don't have any intentional context set up yet, start small: