AI Build Lab / Learn · Agent Workforce · Week 4 · Module 03 · Details

More detail: Pick the problem and agree the fix

Optional reading for this lesson. Everything you must do or decide is in the lesson.

Week 4 · Module 03 · Lesson 1 · Details

This page goes with the Week 4 lesson Pick the problem and agree the fix (course/workforce/week4/lessons/anw-w4-l05.md). You don't need it to finish the lesson. Read it when you want the longer version.

How Evy groups Evy's findings

Evy reads every fail across your cases and groups the ones that went wrong the same way. Each problem says:

  • the cases it shows up in
  • the rules it fails, each with its label and short name
  • where it points: one agent (agent level), or the job level (when every agent did its piece and the team still missed)

Why one problem at a time

The people who study this don't all say "change one thing at a time". What they do agree on is narrower: after every change, rerun the golden set, and keep the change only if the results show it helped.

Fixing one problem at a time is the practical way to do that. If you change things for two problems and a test breaks, you can't tell which change broke it. So all the changes for one problem go in together, and the other problems wait their turn. You fix them later this week, the same way.

What a fix can touch

Kind of changeExample
One agent's instructionsThe writer leaves a gap as a gap
The team's sourcesThe researcher also reads the allergen line in the policies
Kansa's watch listWatch for any diet fact not in the menu guide
The team's designAdd a step, or move a check earlier; a bigger change, which the Architect (Archie) designs with you

How Gigawatt changes an agent's instructions

For a change to one agent's own instructions, Gigawatt uses its Eval & Improve method (aibl-eval-improve-loop): it scores the agent against a fixed scorecard, writes the improved version, scores it again, and keeps the new version only if the score went up. Its notes are saved in the job folder under eval/, one folder per agent.

You approve the plan. The agents do the rest.

You make two decisions about every change to your team:

  1. Approve the plan, in this lesson.
  2. Keep it or undo it, in "Make the fix, rerun the golden set, and keep or undo it".

Everything in between is the agents' job. They stop and ask you only before sending anything outside your organization, posting in public, deleting, moving money, or changing who can get into an account. For anything else, they make a sensible choice, note it for you to check, and keep going.

Changing the plan

Before you approve the plan, you can ask Gigawatt to change anything in it, in your own words. After you approve it, a change to the plan is a new plan, and needs your approval again.