The biggest lever in AI work is context, not a cleverer prompt

Most advice about getting more out of AI is about the prompt. In my experience the far bigger lever is the source material you give it.

Most of the advice about getting more out of AI is about the prompt — phrase it better, add a persona, give it steps. That is not wrong, but in my experience it is the small lever. The big lever is the context you give the model, and most people supply almost none.

Here is what changed how I work. I stopped starting from a blank page. When I write a product spec now, I do not sit down and type it out. I start from the conversations, decisions, examples, and constraints that already exist around the work, and I use AI to turn that pile into a draft I then sharpen. The spec is better, and it is better because the input is better, not because I found a cleverer prompt.

There is a second half to this, and it took me a while to see. When I type into an AI, I clean the thought up before it leaves my head — I cut the doubts, the tangents, the half-reasons. When I speak, all of that messy reasoning comes out. Note that the messy part is the useful part. It is the context the model cannot infer, and typing quietly throws it away. So I talk to the AI far more than I type now, and the output is better for it.

Put those two together and you get the thing Jiva is really about. Working with AI is bottlenecked on how fast and how accurately you can give it your context, and speaking is the fastest way a human does that. So the primary driver behind Jiva was never note-taking, rather building a fast, trustworthy way to get real context — everything you say and hear — into a form any AI can use.

A good model with thin context gives you a confident, generic answer. The same model with the real conversation gives you something specific.

The difference is not the model. It is what you fed it.

More soon.