"Tool use" or "function calling" is one of the most important capabilities in modern AI products. Here's what it means in practice and why it matters for business AI deployments.
Tool use is when an LLM can call external tools — search engines, calculators, databases, APIs — to complete a task. Instead of just producing text, the AI can fetch real-time data, run computations, or take actions in other systems. This is what turns LLMs from chatbots into useful workplace agents.
Plain LLMs can only produce text based on what they were trained on. They can't search the web, can't read your CRM, can't run a SQL query, can't book a meeting. Tool use changes that.
When Claude or ChatGPT has tool use enabled, the model can decide to call a specific tool when it needs information or wants to take an action. Example: you ask "what's our pipeline this quarter?" The model calls a tool that queries your CRM, gets the data, and produces an answer.
Tool use is what makes "AI agents" possible. An agent is essentially an LLM in a loop with access to tools and goals.
Practical workflows enabled by tool use:
Same thing, different name. OpenAI calls it function calling; Anthropic calls it tool use.
No — tool use is opt-in per conversation or per product. Most casual Claude chat doesn't use tools.
Closely related. Plugins were an earlier framing; tools (and the MCP protocol) are the more modern, vendor-agnostic approach.
Yes. Through the Claude API or MCP, you can expose any function as a tool Claude can call.
Agents are LLMs using tools in a loop with goals. Tool use is the underlying capability; agents are the pattern that uses it.