From Words to Actions
An AI agent without tools is like a consultant who can only give advice but can't do anything. It can tell you what should happen, but it can't make it happen. Tool use is what bridges that gap.
Tool use is the process by which an AI agent decides to call a specific function, sends the right information to that function, gets the result back, and uses it to continue the conversation. It's how the agent goes from "I can tell you about your revenue" to "Your revenue last month was $47,200, up 12% from the prior month."
This sounds technical, but the experience from your perspective is simple: you ask a question, and the agent gets you a real answer from your real data.
How It Works: A Step-by-Step Example
Let's walk through exactly what happens when you ask your AI agent a question that requires tool use. We'll use a real scenario: checking your recent sales.
Step 1: You Ask a Question
You: "How much revenue did we bring in last week?"
The agent reads your question and recognizes that answering it requires real data from your payment system. It can't just guess -- it needs to actually check.
Step 2: The Agent Chooses a Tool
The agent looks at the tools available to it. It's connected to Stripe, so it has access to tools like "list charges," "get balance," and "search transactions." It decides that searching recent transactions is the best way to answer your question.
This decision happens automatically. The agent understands what you're asking and picks the right tool, just like you'd know to open Stripe (not your email) to answer a revenue question.
Step 3: The Agent Calls the Tool
The agent sends a request: "Get all successful charges from the last 7 days." This request goes to your Stripe integration through Pipeworks, which handles the connection securely.
Step 4: The Tool Returns Results
Stripe sends back the data -- a list of charges with amounts, dates, and customer details. This raw data goes back to the agent.
Step 5: The Agent Interprets and Responds
The agent takes the raw data and turns it into a useful answer:
Agent: "Last week, you brought in $23,400 across 47 transactions. That's up 8% from the previous week. Your largest single transaction was $2,800 from Acme Corp on Tuesday."
That's it. Five steps, all happening in a few seconds, completely behind the scenes.
A More Complex Example
Tool use gets more interesting when the agent chains multiple tools together. Here's what that looks like:
You: "Compare our website traffic to our revenue for the past month. Are we getting more visitors but fewer sales?"
What the Agent Does
- Calls Google Analytics tools -- Pulls website traffic data for the past 30 days (sessions, page views, unique visitors)
- Calls Stripe tools -- Pulls revenue data for the same period (total charges, transaction count, average order value)
- Analyzes both datasets -- Compares the trends, calculates conversion rates, identifies patterns
- Responds with insight -- "Your traffic is up 15% but revenue is flat. Your conversion rate dropped from 3.2% to 2.7%. This suggests visitors are finding your site but something is preventing them from purchasing."
The agent called tools from two different integrations, combined the results, and gave you an insight that would have taken 20 minutes of manual analysis.
Tool Permissions: Controlling What the Agent Can Do
Every tool falls into one of three categories:
Read Tools
These retrieve information without changing anything. Looking up a customer, pulling a report, checking a balance -- these are all read operations. They're completely safe because they can't modify your data.
Write Tools
These create or update records. Sending an email, creating an invoice, updating a contact -- these make changes to your systems. They're useful but should be granted deliberately.
Delete Tools
These remove records permanently. Canceling a subscription, deleting a contact, removing an invoice -- these are the most powerful tools and should only be enabled when you specifically need them.
You control which categories of tools your agent can access. Most businesses start with read-only access and add write tools once they're comfortable with how the agent works.
What Happens If a Tool Fails?
Tools can fail for the same reasons any software can -- the service might be temporarily unavailable, your credentials might have expired, or the request might contain invalid data.
When a tool fails, the agent doesn't crash or do something unpredictable. It gets an error message, understands what went wrong, and tells you:
Agent: "I wasn't able to pull your Stripe data -- it looks like the connection needs to be refreshed. You can reconnect it in your Pipeworks dashboard."
The agent handles failures gracefully, just like you'd handle getting an error message when you try to log into a website.
Why This Matters
Understanding tool use helps you:
- Trust the process -- You know the agent isn't guessing or making things up. It's pulling real data from your real systems.
- Set the right permissions -- You can make informed decisions about which tools to enable and which to keep turned off.
- Troubleshoot issues -- If an answer seems wrong, you can check whether the right tool was called and whether it returned the expected data.
- Get more out of your agent -- Once you understand what tools are available, you can ask more specific and powerful questions.
You don't need to tell your agent which tools to use. Just ask your question naturally, and the agent figures out which tools to call. It's designed to be as simple as talking to a knowledgeable assistant.
The Human in the Loop
Tool use doesn't mean the AI works without oversight. You're always in control:
- You choose which integrations to connect
- You set permission levels for each integration
- You can review what tools were called in the activity log
- You can disable specific tools at any time
The agent is a powerful assistant, but you're the one making the decisions about what it has access to and what it's allowed to do.
Getting Started
Connect your first integration and ask your AI agent a simple question about your business data. Watch how it uses tools to get you a real answer in seconds -- that's tool use in action.