Module 01 Activity

Scenario

A colleague has proposed a five-agent crew for a workflow at your organisation. Before any code is written, the proposal has to survive the three-question test.

What you build

A crew design reduced to the smallest set of agents that could work, with the reduction argued.

Steps

1. Pick a real workflow. Write out every step as a verb phrase.
2. Write a first-pass agent list - one per step, deliberately over-split.
3. Score each against different tools, different permissions, different judgement. Merge anything clearing fewer than two.
4. Keep any agent that is the only one with write access, regardless of its score, and say why.
5. Write expected_output for every remaining task so a reviewer could reject a submission against it.
6. Build the reduced crew and print its agent count, task count and process.

Evidence to hand in

• The step list and the deliberately over-split first pass.
• The three-question scoring table with merge decisions.
• The final crew with expected_output per task.
• One sentence per merged agent saying what was lost.

Review checklist

• The first pass is genuinely over-split, so the reduction has something to do.
• Every merge decision cites which bars the agent failed.
• Any write-access agent is kept separate with a stated reason.
• No expected_output could be satisfied by 'a good result'.
• The final count is lower than the first pass, or the reason it is not is written down.

Mark complete