Keep Your Own Personal Context

The real value of AI/LLM's isn’t comes not just in using them, but in making sure you set yourself up to use them well.

We're all still early in this new AI world, but I can already see my own workflow has changed. Codex is my main tool at the moment, and I'm increasingly treating it less like a useful knowledge assistant and more of a main interface for my daily work.

That doesn’t mean handing everything over to it and never checking what it does. It means learning where AI is genuinely useful, reduces friction, and where it creates more value than complexity, while trying to avoid the rabbit holes - regular course correction!

The more I use these tools, the more I think the best thing you can do is keep your own context in the simplest possible format and store it independent of the tool itself. Notes, tasks, plans, project context, decisions, drafts, loose thinking. All of it becomes more useful if it’s stored in a way that is clear, portable, and easy to move between tools. Right now, I'm using Notion for this, but a series of (.md) files would work just as well if I'm being honest!

In practice, this mainly takes the form ask Codex to write out decisions/input/notes to Notion as well as to memory, so I can take advantage of the built-in tools while still keeping my personal context portable.

Keeping a copy of your own context matters because as the tooling is changing so quickly, going all in on one tool is next to impossible, especially if you work across disciplines. One day you’re designing an image in ChatGPT. The next you’re using Codex for a Heroku upgrade. Then Claude for a writing project... Some of them are better at writing, some at planning, some at code, some at handling longer context. If your system depends too heavily on one platform’s structure, you'll end up locked in at a time when it's genuinely hard to pick a winner or find the right tool(s) that work best for you.

What seems more durable is to treat the tools as interchangeable layers on top of your own context, rather than treating the tool itself as the system. The value is in making sure the underlying material is yours, legible, and easy to reuse. Your individual context is the key.

That means simple notes. Clear task definitions and logging decisions. Project context that can be lifted and dropped into a prompt without much effort. Enough structure to be useful, but not so much structure that maintaining it becomes its own job.

The tools are going to keep changing. They’ll get better, more specialised, more integrated, and probably more fragmented too. So the advantage, at least from where I see it, is not trying to predict the winning tool. It’s building a way of working where you can switch between them as needed, and using the right tool for the job.