Showing work to the person on the AI Agents track. People trust an agent when they can see actions before harm and understand why a step happened. Hide the raw chain of thought if you want, but do not hide tool calls that change data.
This lesson assumes you already worked through What the agent is allowed to see.
The idea in practice
Show a running list: searched, found, waiting for approval. Afterward, show what changed. Offer undo where the tool supports it.
A concrete check
goal = {
'track': 'AI Agents',
'lesson': 'Showing work to the person',
}
checks = [
'input available at decision time',
'score matches the real decision',
'failure case written down',
]
print(goal['lesson'])
for item in checks:
print('-', item)
Run the sketch locally if you have Python. The printout is a reminder of the checks, not a trained model. Replace the strings with the real inputs from your own example before you treat it as a design.
What usually goes wrong
A silent agent that 'just handles it' produces support tickets you cannot explain. When this happens, stop adding parameters or tools. Fix the check, the data, or the permission, then run the same example again.
What to write down
- The input you are allowed to use at decision time.
- The output and the score or pass rule.
- One failure you will test on purpose.
- What you will not claim the system can do.
Practice
Sketch the three UI states for a refund: proposed, approved, done.
Self-check
- Say Showing work to the person in one sentence that mentions an input and an output.
- Name the failure mode in this lesson and the check that would catch it.
Done when: you can explain this lesson without the page open, and you have a written failure case.