Tool calling, integration and automation
An integration connects two systems.
An automation always runs the same steps when a condition is met. In tool calling it is the model that decides, in each conversation, whether a function is needed and with what data. That is why it needs limits the other two do not.
Why it matters
It is what turns a text generator into something that resolves.
Without tools, the most a system can do is explain how to raise a limit; with them, it raises it and confirms the result in the same conversation. It is the piece that separates a chatbot from an agent.
How it works
The model gets a description of each tool.
What it does, what parameters it needs and what it returns. Faced with a question, it either answers directly or asks for a tool to be run. Your system runs it, returns the result, and the model writes with it.
The model does not run anything
An important nuance for understanding the risk.
The model only asks for a function to be called. Your system is what runs it, and it can check permissions, validate arguments and refuse. That is where the limits live, not in the model.
Which tools to expose
The useful rule is the same as with any permission.
Only what is needed, and reads separate from writes. Checking a plan is a read; canceling a subscription is a write, and probably should not be available without a person confirming.
Examples in customer service
Read-only ones come first.
The ones that change something are added later and, when they cannot be undone, with a person confirming.
- Checking the status of an order or an invoice.
- Looking up the plan and usage of an account.
- Resending a verification email.
- Changing a contact detail.
- Opening an issue for product with the case already described.
