Start Thinking About Your Context

I never wrote anything down. Not really. Every idea I had, every view I formed, every conclusion I reached, lived in my head or surfaced once in a conversation and then disappeared. For most of my career that was fine. Then I started using AI seriously and realised it was the problem.

The first thing that nudged me was follow-through. I'm bad at it. I bought a Plaude to help me start capturing the meetings and the thinking I was doing out loud, because I knew I wouldn't reliably sit down and write any of it down myself. That was a start. But then I learnt about agentic, and that's when the whole thing clicked.

The click was simple. To get anything genuinely useful out of an AI, I had to bring the why with the what. Telling it what I wanted produced vague output. Telling it what I was working on, why it mattered, what I'd already tried, and then asking for a suitable outcome, that's when the results stopped feeling like search engine results and started feeling like work I would have done myself if I had the time.

And once I'd seen that I couldn't unsee it. Every prompt I wrote without that context was leaving value on the floor. Every conversation where I had to re-explain who I was and what I was building was a tax on my own thinking. The bottleneck wasn't the AI. It was me, and the fact that almost nothing I knew was in a form the AI could read.

Capture once, not every time

So my goal shifted. Not "how do I write better prompts." That's the wrong question. The right question is how do I capture as much as possible with as little additional effort as possible, so the AI already has the context before I open my mouth.

That meant a few things. A Plaude for recording, transcribing, summarising, sometimes analysing the meetings I'm in. Skills for Claude that write things down automatically as we work, so the conversation doesn't end in chat memory that I'll lose. Private emails and documents connected in. A scheduled sweep every three hours that picks up what's accumulated and turns it into context I can actually use. None of it requires me to remember to log anything. The infrastructure runs in the background while I live a normal day.

One thing I learnt fast. The context has to live somewhere that isn't tied to any single AI provider. Models leapfrog each other weekly now. Claude leads one week, Gemini the next, something else after that. And the right move isn't to pick a winner. It's to use whichever model is best for whichever task, and let your context be the constant. Your context is the asset. The models are interchangeable engines pointed at it.

Where this is going

The future I'm building toward, and I think we're closer to it than people realise, looks like this. I had a thought in the shower. I wrote it down, asked Claude to work on it, and by the time I was dressed and sat at my computer there was a mockup of an app ready for me, and it was already suggesting features to refine and add.

That wasn't because the AI was clever. It was because Claude already knew enough about me, my taste, my stack, how I think about products, that "build this" was a complete instruction. Without that context, I'd still have been back-and-forthing for an hour just to get the scaffolding right.

Throwaway ideas stop being throwaway. They turn into something you can look at and decide whether to pursue, with almost no friction between thought and prototype. The improvement on an existing thing you've already built can happen in the time it takes to make a coffee. That's what the context is buying you.

You can't backfill this

You can't backfill this. Try it. Pick a decision you made two months ago and write up the reasoning that led to it. You'll get the bones. What you decided, roughly when, roughly why. What you won't get is the texture. The shower-thought that actually drove it. The half-sentence in a meeting that reframed your thinking. The gut reaction you had before you rationalised your way into a position. That's the layer the AI actually needs, and it's already gone.

The valuable context isn't the conclusion. It's the route you took to get there. And the route evaporates within days.

That's why this is urgent. Not because you'll be slower for a year if you wait. Because the most valuable context has a shelf life of days, and every day you don't capture it is a day lost permanently.

Self-host it

One more thing, and this is the part I'd flag the hardest. Self-host it. All of it. Your context will become one of the most sensitive things in your life. It will know your secrets, your patterns, your money, your relationships, the things you're proud of and the things you're not. Keeping it in your personal domain is so, so important.

It's not even that hard if you only use a phone and a laptop. It gets more complicated the more tech you interact with. But the time to protect it is from the start. Once it's in someone else's platform that ship has sailed.

Where this gets uncomfortable

I'll be honest about a few things I haven't solved.

Running this privately means running it on consumer-grade hardware, which is the best balance I've found against spending way too much, and that hardware caps how good the local model can be. It misses things, and retrieval can be slow. The genuinely private option works, but it doesn't yet work as well as I'd like.

Then there's the part that actually keeps me up. The more you capture, the more your context becomes a concentration of highly sensitive data, and that makes it a target. Prompt injection is still a real threat, and that's before you even get to the security of the setup itself. You can build safeguards to screen what leaves your domain, and I have, but they narrow the risk, they don't remove it. Self-hosting is the right instinct, but it moves the security problem onto you, and that is not a small thing to take on.

And capturing everything is not the same as capturing well. A store full of noise is its own kind of problem, though I currently use Open Notebook to map the context, which manages that reasonably well for now.

None of this is a reason not to start. But this isn't a solved, clean practice, and I'd be wary of anyone who tells you it is.

Start now. Even badly. A phone notes app and a habit of dumping the day before bed beats nothing, and nothing is what most people have. The tools you use will change. The model you use will change. The context you start capturing today is the only thing that survives the next round of upgrades, and the only thing that compounds while you sleep.