Tool Result
What comes back after a tool runs — the data or message the model reads to decide its next move.
Think of it like
The number the calculator shows after you press equals; you read it, then keep going.
Example
The weather tool returns `{temp: 14, condition: "rain"}`; the model reads it and writes “Bring an umbrella — it’s 14° and raining.”
How it actually works
The tool result is appended to the context as the fruit of an action, and the model’s next turn is conditioned on it. Its format matters more than people expect: a clean, relevant result guides the model well, while a giant raw dump wastes context and buries the signal. Errors count as results too — a good agent reads a failure and adapts rather than plowing ahead.
For product teams
Feed agents tidy results, not fire-hoses — how you shape tool output shapes the answer.
For engineers
The output of a tool execution appended to context, conditioning the model’s subsequent generation.
Related
- Tool Calling — The action that produced it.
- Observation — Its role in the perceive-then-act loop.
- Agent Loop — The loop it feeds back into.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome