How I work

Build first, talk later

Most software projects start by talking about the software. I start by building it.

Talking first is the expensive part

A spec describes something nobody has seen yet. The meetings about it are guesses, and the estimates are guesses about the guesses. People are good at reacting to something they can use, and bad at imagining it. So the first weeks of a normal project go into producing agreement about a thing that does not exist.

So I move the building to the front

You describe the idea in a few lines. I build a version that works. Then we look at the real thing and decide what happens next. The first version is the discussion, not what comes after it.

Building used to be the slowest and most expensive step, which is why everyone tried to get it right on paper first. That order made sense when a first version cost weeks. It costs days now. When building is the cheap part, it is also the fastest way to find out what you actually want.

daysweeks

  • Talk first

    spec, meetingsbuild

  • Build first

    buildrework

Build first wins. The build is cheap enough that finding out from the real thing costs less than agreeing on paper first.

Drag it and the two bars cross. Rework is the share of a build you lose when the first version turns out to be the wrong thing, which is exactly what a spec is bought to prevent. A model rather than a measurement: the bars carry proportions, not hours.

In practice

AI makes the building fast. It does not decide anything. I review the work, I choose what goes in, and I am responsible for what ships.

No spec needed to start. A working version in days. You use it, you react to it, and we keep it, change it, or drop it. Different problems need different technologies, and I pick what fits the problem instead of reusing one stack for everything.

  • A tool for comparing places to live, side by side.

  • A planning board for a trip with a lot of moving parts.

  • A pipeline that keeps a large catalogue in order.

  • MDP, a compiler that turns written notes into pages, decks and flyers.
  • entailer, a toolkit for checking whether an argument actually follows.
  • temporal-plugin, durable workflows for agents, as a plugin others install.

See it for yourself

A small game, built in an afternoon because that was faster than describing it.

The landing page has the work I have built this way, and a short form to describe an idea.

See what I build