blog :: Working notes from whilst doing the work.

Where is the low hanging fruit with AI? - Start with the boring work. Not the clever work. The work that is done the same way every time, because that is where you can tell whether it was done right.

An AI agent charges you per answer and never says whether the answer was any good - Telemetry records the cost, the template, the model and the latency. It does not record whether the answer was any good, and that half has to be built by hand.

The cheapest way to build an AI agent is the most expensive way to run one - The option that feels like progress in week one is the one still charging you in week fifty. An agent has to keep its policy somewhere. There are three obvious places to put it.

Interesting to me. Not to you. - What stops you pressing the button? Gut feel, a friend, a chat AI, market research.

Your next vibeCode starts where the last one finished - The first vibeCode is rough. It is also the worst it will ever be. What the next one is worth depends entirely on what you rebase it onto.

I wrote down everything I know about building with AI - A year of building, written down. The agent loop, why more tools make an agent worse and what a gate should actually check.

Thank you and a clearer vibe2value - Your feedback to yesterdays email resulted in some changes. The big one: the site now says plainly what it is, a way of thinking about building with AI.