Skip to main content
New: 9 PM Courses with hands-on exercises and certificates
Back to Glossary
AI and Machine LearningT

Tool Use

Definition

Tool use (also called function calling) is the capability that allows language models to interact with external systems by generating structured requests for specific functions or APIs. Without tool use, an LLM can only generate text based on its training data. With tool use, it can query databases, call APIs, execute code, send messages, and perform any operation exposed as a callable function.

The mechanism works through a structured contract. The developer defines available tools using JSON Schema descriptions (name, description, required parameters, optional parameters). When the model encounters a request that would benefit from a tool, it generates a structured JSON object specifying which tool to call and with what arguments. The application layer validates the request, executes the tool, and feeds the result back to the model. The model then incorporates the result into its response.

Tool use is the foundation of AI agents. A model that can call tools in sequence, evaluate intermediate results, and decide what to do next is fundamentally an agent. The Model Context Protocol (MCP) is an emerging standard that makes tool definitions portable across model providers, reducing the integration burden for products that support multiple LLMs. You can explore how tool-augmented AI connects to your product strategy with the AI Build vs Buy Assessment.

Why It Matters for Product Managers

Tool use is what makes AI features genuinely useful rather than merely conversational. A chatbot that can look up a customer's account status, check inventory, or create a support ticket delivers real value. One that can only discuss these topics from its training data does not. For PMs, tool use is the bridge between "AI that talks about your product" and "AI that works within your product."

Designing the tool set is a PM responsibility, not just an engineering one. Which actions should the AI be able to take? What data should it access? What permissions should it have? These decisions shape the user experience and the risk profile of your AI feature. A customer support agent with access to refund processing, account modification, and escalation routing is a fundamentally different product from one that can only read account data.

How to Apply It

Start by mapping user intents to the tools needed to fulfill them. Every workflow your AI feature supports needs a matching tool set.

Steps for implementing tool use:

  • Audit existing APIs and internal services that could be exposed as tools
  • Write clear tool descriptions (the model's ability to choose the right tool depends on description quality)
  • Define parameter schemas with validation constraints (required fields, enum values, min/max)
  • Implement a permission layer: which users and contexts allow which tool calls
  • Add guardrails for destructive operations (deletes, sends, payments) with confirmation steps
  • Log every tool call with inputs, outputs, and the model's reasoning for audit and debugging
  • Test tool selection accuracy: does the model pick the right tool for ambiguous requests?

Frequently Asked Questions

How does tool use work in language models?+
The developer provides the model with a list of available tools, each described by a name, purpose, and parameter schema (usually JSON Schema). When the model determines that a tool would help answer the user's request, it outputs a structured tool call instead of a text response. The application executes the tool call, returns the result to the model, and the model uses that result to generate its final response. The model never executes tools directly. The application layer mediates all tool calls.
What is the difference between tool use and function calling?+
They are essentially the same concept with different naming conventions. OpenAI initially called the feature 'function calling' and later renamed it to 'tool use' in their Assistants API. Anthropic calls it 'tool use' in Claude. Google calls it 'function calling' in Gemini. The mechanics are identical: the model outputs a structured request, the application runs it, and returns the result. In practice, both terms are used interchangeably across the industry.
What types of tools can AI models call?+
Any function exposed through an API or callable interface. Common categories include: data retrieval (database queries, search APIs, CRM lookups), computation (calculators, data analysis scripts, code execution), communication (sending emails, Slack messages, creating tickets), file operations (reading documents, generating spreadsheets, creating PDFs), and external services (weather, maps, payment processing). The only constraint is that the tool must accept structured input and return structured output.

Explore More PM Terms

Browse our complete glossary of 100+ product management terms.