Can the model invent a tool result?

It can describe a result it never received, which is why the software, not the model, must run the tool and pass the raw result back. Show the tool's answer in the log. If the reply mentions an order status that the tracking tool did not return, that is a bug.

Who decides which tools an AI system can use?

You do, in the design, not the model in the moment. List the tools, the fields each one accepts, and whether it reads or writes. Review that list the way you would review a new employee's system access. Add a write tool only after the read-only version has been right.

Agents & automation

What is function calling in AI?

It is what connects an AI model to your actual systems. Without it, an assistant can describe an action; with it, the assistant can perform one.

In plain terms

Tool use, also called function calling, is an AI model's ability to trigger real actions, like querying a database or sending an email, instead of only producing text.

Also called Function Calling.

Function calling, also called tool use, is how a model uses a real system instead of guessing. You describe a tool, such as look up this order number. The model asks for that tool. Your software runs it and gives back the result. The model then writes from the result.

This is the line between a chatbot and something that can change the business. Read access, such as looking up an order, is a reasonable start. Write access, such as refunding it, needs a permission, a log, and usually a person. The model should only see tools you listed, not a general key to the company.

Think of it this way: An LLM without tool use is a brilliant advisor locked in a room. Tool use is the telephone, the keyboard, and the filing cabinet that lets the advisor act on their advice.

A customer-facing assistant uses tool use to call a live inventory API, check real-time stock levels, and confirm a delivery estimate in the same conversation, without any human lookup in the middle.

A buyer asks where my shipment is. The model calls the tracking tool with the order number and answers with the carrier status. It cannot call the refund tool. That tool is not on its list. A person handles refunds, and the lookup still removes a large share of status calls.

Whenever the AI needs to take action beyond producing text: querying databases, updating records, sending notifications, or calling third-party APIs. Tool use requires careful permission design. Do not grant an AI access to irreversible or sensitive actions without a human confirmation step at those specific decision points.

RaftLabs builds agents that do one defined job, with a person still on the steps that cannot be undone. You see the workflow on your own data before it runs on its own. The related work on our side is AI agent development.

This sits with the other agents & automation terms on the glossary. When software stops only answering and starts carrying out the work. Worth reading next: AI Agent, Agentic AI, and Multi-agent System.

Common questions

: It can describe a result it never received, which is why the software, not the model, must run the tool and pass the raw result back. Show the tool's answer in the log. If the reply mentions an order status that the tracking tool did not return, that is a bug.
: You do, in the design, not the model in the moment. List the tools, the fields each one accepts, and whether it reads or writes. Review that list the way you would review a new employee's system access. Add a write tool only after the read-only version has been right.

Work with us

Tell us what's broken.

Tell us what's not working in your business. We'll find the real problem and tell you exactly what it would take to fix it.

Scope and cost agreed before work starts. No surprises. No obligation.
Working prototype within 3 weeks of kickoff.
Pay by milestone. You see progress before each invoice.
60-day post-launch warranty. Bug fixes, UI tweaks, and deployment support. No retainer.
All conversations are NDA-protected.