Use this whenA team has a real workflow worth improving and you can point an AI developer at it for a few weeks.

Why it exists

Interviewing a team about its work gets you the tidy version. Building from a requirements document gets you a tool that arrives months later and misses the point. A developer who sits with the team sees the real work, and a team that watches the build trusts the result.

The framework

  1. Step 1Sit inWhat does the work actually look like?
  2. Step 2Map one piece of workWhat should change first?
  3. Step 3Build in the openWhat is the smallest useful automation?
  4. Step 4Hand it overWho owns it when you leave?
  1. Sit in

    What does the work actually look like?

    Watch the team do the real work, rather than only interviewing them. Note the waiting, the re-keying, and the workarounds.

  2. Map one piece of work

    What should change first?

    Map one piece of work with the team, using data, process, people. Choose a step worth removing or automating.

  3. Build in the open

    What is the smallest useful automation?

    Build it in days, with the team watching and correcting. Show it early and often, and use real work as the test.

  4. Hand it over

    Who owns it when you leave?

    Name an owner, add it to the tool registry, leave a champion behind, and move on to the next team.

How to use it

  1. Time-box it to a few weeks per team, and say so on day one.
  2. Spend the first days watching, then map one piece of work together.
  3. Build the smallest useful automation in days, in front of the team.
  4. Test it on real work with the people who do it.
  5. Hand it over with an owner and a registry entry, find or grow a champion, and leave.

What it looks like

Picture a contracts team that spends part of every week copying terms out of documents into a tracker. A developer sits with them, watches the copying, and builds a small tool that drafts the entries for a person to check. By the end of the stay, the team lead owns it, it has a registry entry, and one person on the team knows how to adjust it. The developer moves on to the next team.

Where it breaks down

The developer can become the team’s permanent support desk, so the handover needs a firm date. Developers drift toward what is interesting rather than what is valuable, so keep returning to the mapped workflow. And a team with no champion lets the tool die. If you can’t find one, that is a reason to wait, not to build.

Try it with AI

I'm going to send an AI developer to embed with a team for a few weeks. Help me plan it.

Ask me one question at a time about the team, the workflow, and the people who do it. Then outline: what to watch for while sitting in, how to map one piece of work using data, process, and people, what the smallest useful automation might be, and what a clean handover needs (a named owner, a registry entry, and a champion).

Finally list the ways this could go wrong, such as becoming a permanent support desk, and how to guard against each.

Related