Services

AI features

Useful AI is a feature with a job, not a chat window bolted to the side. We add language models where they save a person real effort, and we ship each one with limits, fallbacks and a number that says whether it is helping.

Lines of code on a screen, seen at an angle

What you end up with

  • Drafting and planning tools built into your own screens
  • Recommendations that explain why something was picked
  • Internal agents that take over repetitive back-office work
  • Spending limits and caching so costs stay predictable
  • A plain fallback for when the model is slow or wrong

How we do it

  • Claude API
  • Tool use and agents
  • Prompt caching
  • Evaluations
  • Python
  • Start with the task

    In Milau, 'Plan my weekend' and 'Start something' exist because people froze at an empty form. The model fills the form; the person stays in charge.

  • Measured against doing nothing

    Every AI feature has a switch and an event trail, so its effect can be compared with the version without it.

  • Private by design

    Only what the task needs is sent to a model. Nothing is used to train one.

Questions about this

Which model do you use?

Claude by default. The integration is written so the model can be changed without rewriting the feature.

Will it make things up?

It can, so we design for it: sources shown, actions confirmed by a person, and checks on what comes back.

Tell us what you are making.

A few lines is enough. A person reads it and writes back with what it would take.