Tool Use

Tool use is the LLM's ability to call external functions — search a database, send an email, run code, hit an API — instead of only generating text.

Plain LLMs only output text. Tool use turns them into actors. The model is given a list of available tools with schemas; when it decides it needs one, it emits a structured call; the system runs the tool and feeds the result back into the model. The loop is the heart of agentic AI. Tool use unlocks everything that requires real-world side effects: looking up live data, modifying state, integrating with SaaS, reading files. Without tool use, an AI can talk about doing a job. With tool use, it can do it.

Example

AI EA receives 'book me a flight to NYC Tuesday'. It calls search_flights, picks an option, calls book_flight with the user's payment token, calls send_confirmation. Five tool calls, one task.

How OpenLabor uses it

OpenLabor connects employees to your tools via MCP and direct integrations. New tools register without redeploys.

Is 'function calling' the same as tool use?

Yes — different vendors use different names for the same capability.

How do I keep a model from calling tools wrong?

Clear tool descriptions, strict schemas, validation on the call before execution, and an approval gate on irreversible tools.

Related: mcp, agentic-workflow, guardrails.

AI Labor Glossary