New: Cookbooks and AI ExplanationsStep-by-Step recipes to solve problems connected to Roadmaps and Cheat Sheets. Need more details? Use AI buttons for structured and simple explanations with concrete examples throughout the whole platform.Take a look
The output looks fine. There's no way to know if your agent is quietly wrong.
What you'll have at the end
A verification pass that checks the agent's answer against the original goal before it's returned, with a second try triggered on a fail
You need
A tool-using agent that already returns a fluent, complete-looking answer straight to the user after calling whatever tools it needed, with nothing standing between the last tool call finishing and the answer being sent.
Not covered
Deciding whether a risky tool call needs a person's approval before it fires; this recipe only grades the finished answer after every tool call has already run.
Leans on
Should your agent ask before it deletes, sends, or pays?
That recipe gates a risky action before it runs; this one checks the finished answer after every action has already completed, so use both together for an agent that both acts and reports back.
Log every tool call so you can find what broke a run
That recipe logs every tool call across a whole run; this one's audit record only covers the verification decision itself, so pair them for full coverage of what happened and why the final answer was trusted.
Write a tool-using agent loop that knows when to stop
If the agent doesn't yet have a clean loop that knows when it's done, build that first: a verification pass needs one settled answer to grade, and a loop still deciding whether to continue never hands it one.
Checked 15 Aug 2026
Part of the AI Agents cookbook