9.3 Tools, function calling and action spaces
You can give a model capabilities safely.
Before:08. Large Language Models
Tools are how a model acts on the world, and schema design is most of the game: names and descriptions the model interprets correctly, arguments it can fill reliably, idempotency and retries deciding what happens when calls fail mid-task. It follows the loop topic because loops act through tools. The bright line is drawn early: irreversible actions get human confirmation, because wiring delete or send or pay directly to a model's judgement is how demo stories become incident reports.
Work through these
Designing a tool schema the model uses correctly
A tool description is a prompt: the name, the parameters and the description all decide whether the model uses it correctly. Most tool failures are description failures.
Tool selection failures and disambiguation
Models pick the wrong tool when two sound similar, and the fix is usually in the naming rather than in the model. Making tools distinguishable is a design task.
Idempotency, retries and side effects
A tool that can be safely called twice, retried on failure, and whose effects are understood. Without this the retry logic above becomes dangerous.
Human-in-the-loop confirmation for irreversible actions
Anything that cannot be undone should require a person to approve it. This is the single most important safety control in an agent that acts on the world.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.