CourseAgentic AI Engineer: The Complete Course · Module 11: Multi-Agent Systems · part 63 of 97
Part 63 · Module 11: Multi-Agent Systems

Lesson 3: How agents hand work to each other safely

27 min read·13 Oct 2026
Module 11 · Lesson 3 · About 60 minutes · Needs: the shared data's compact tools (make_tools, run_tool; Module 3 builds their final form) and Module 5, Lesson 1 (the loop and ScriptedModel)

The situation

It is Wednesday 3 June 2026, week 8. Dev has prototyped his split on the 20 April queue: a lead agent hands R-1001, Aisha Rahman's robot vacuum (order 5521, $129.00), to an eligibility checker and then to a case worker.

It half works. Priya finds two returns for one vacuum: R-1006, Aisha's duplicate, ran at the same moment. Marta asks which agent spent what. Sam asks what stops one agent spawning a hundred. Whatever docs/multi-agent-decision.md concludes, any split must get these mechanics right.

R-1001, SPLIT THREE WAYS lead plans, decides eligibility-checker 2 read tools case-worker 3 write tools BrightCart systems (shared)
The prototype: a lead (blue, your code's coordinator) hands Aisha's return to two subagents (amber, model-driven), and both touch the same shared systems.

The rest of this course is yours to keep

This course is bought on its own, once, and stays readable afterwards, including the parts added to it later.