Lesson 166 · 2 min
When AI teams fail
Teams of agents fail in very human ways: vague briefs, lost messages, duplicated work and nobody checking the final result.
The most common failure is a bad handoff. The lead writes "research the market" and the helper has no idea which market, which country or how deep to go. It returns something long and useless, and the lead accepts it.
Errors also spread. If one helper invents a figure, the next helper builds on it, and the final report repeats it with confidence. Each step looks reasonable while the whole chain is wrong.
Teams can loop or stall too. Two agents keep passing a task back and forth, or a helper waits for an answer that never comes. Meanwhile the bill for AI time keeps growing.
Fixes are the same as for people: clear briefs with a goal, limits and the format to return; a checker step that compares claims with sources; a cap on steps and spending; and a person who reviews the result before it is used.
Do it now
Test a handoff
- 1. Write a one-line brief you might give a helper, like "find good suppliers".
- 2. Ask a chatbot to point out what is missing, with the prompt below.
- 3. Rewrite the brief with goal, limits, sources and the output format.
Here is a brief for an AI helper: "[YOUR BRIEF]". List five questions the helper would need answered to do this well. Then rewrite the brief so it answers them.
Quick check
Words from this lesson
- Agent
- An AI system that takes several steps on its own, such as searching, clicking or using apps, to finish a goal you set.
- Hallucination
- When an AI states something false with confidence, such as a made-up fact, quote or source.
- Human in the loop
- Designing a process so a person reviews or approves the AI's work before it has real consequences.
- Eval
- A test you design to check whether an AI system does your specific job well, using examples with known good answers.