← Meet AI · What is an AI Agent?
Learning path · Step ④
How do Agents use tools?
Agents don’t only generate text—they call tools to search, read files, run code, hit APIs, and finish real work.
In simple terms:The LLM thinks, tools act, and the Agent decides when to use which tool.
- You
- "Research the AI agent market"
- Agent
- 🧠 Think: I need fresh sources
- 🔎 Web Search
- 🧠 Analyze results
- 📄 Read sources
- 🧠 Organize findings
- 📝 Deliver a report
Why do Agents need tools?
Without tools, an LLM mostly Understands → generates
Explain what an AI Agent is.
It can answer directly.
Find today’s latest AI Agent news.
🔎 Search — for up-to-date information.
Analyze this PDF.
📄 File tools — to read the document.
Write a Python script and run it.
💻 Code execution — to run and verify.
Tools turn AI from “knowing how to answer” into “getting data and taking action.”
What tools can Agents use?
- Web Search
Search the open web.
- Latest sources
- News research
- Market scans
- Competitive intel
- Files / PDF
Read and analyze documents.
- PDFs
- Docs
- Reports
- Data files
- Code
Run code and work with data.
- Python
- Analysis
- Charts
- File transforms
- API
Connect to external services.
- Weather API
- Agent
- Get forecast
- Database
- Agent
- Query rows
- MCP
An open way to plug in more tools and data.
What is MCP →
How does an Agent pick a tool?
It’s not always a fixed script—the Agent reasons about the task.
Research AI startups in Tokyo.
- Task
- 🧠 Agent analyzes
- Need fresh info?
- Yes
- Use Web Search
- Results returned
- Need more?
- Yes → search again
- Analyze
- Report
Agents choose when—not just whether—to use tools.
What is Tool Calling?
Tool Calling — Tool calling is how an LLM asks an external tool to perform an action and then uses the result.
- LLM
- "I need to search"
- Tool call
- Search tool
- Returns data
- LLM continues
Example
- User:
- "What's the weather in Tokyo now?"
- Agent / LLM
- "Need weather data"
- Call Weather Tool
- Tool returns data
- Agent / LLM interprets
- Reply to user
The tool itself need not be AI—a search API, database, calculator, or script can all be tools.
The full tool loop
- User goal
- Understand task
- Plan next step
- Pick a tool
- Call tool
- Get result
- Analyze
- Decide next step
- Call again if needed
- Finish
🔎 Research Agent
- Understand task
- Plan research
- Search the market
- Read sources
- Compare companies
- Analyze trends
- Organize notes
- Write report
In progress
Research the AI agent market.
- Understand the task
- Create a plan
- Search sources
- Read materials
- Analyze the market
- Draft the report
Can one Agent use many tools?
Yes—and that’s where Agents get powerful.
- Research Agent
- 🔎 Search 📄 Files 💻 Code
- 🧠 LLM
- 📝 Report
- CSV file
- File tool
- Code tool
- Analysis
- Charts
- Report
Complex Agents orchestrate several tools—not just one.
Will Agents misuse tools?
Good systems put guardrails in place:
- Permissions
What can the Agent reach—files, DB, internet?
- Parameters
What inputs is each tool allowed to accept?
- Scope
Read-only vs modify vs delete—explicitly bounded.
- Agent
- Tool
- Permission check
- Allow → run
- Deny → stop
“I’m about to send this email—please confirm.”
This leads to permissions and human-in-the-loop on serious actions.
Tool vs Agent
| Tool | Agent |
|---|---|
| One capability | One goal |
| Waits to be called | Plans the next step |
| Search, calc, read file | Research, write, build |
| A component | The task system |
| “Do this action” | “Help me finish this” |
🔎 Search tool: run this query.
🧠 Research Agent: research this topic.
- Search → read → search again → compare → analyze → report
Tools are abilities; Agents orchestrate them toward an outcome.
How does MCP relate to Agent tools?
- Agent
- Tool
- MCP
- External tools / data / services
MCP standardizes how Agents discover and connect external tools and data—without a custom plugin per app.
Why Agent + tools matter
Before
- You
- Find information
- Operate apps
- Call services
- Compile results
Now
- You
- State a goal
- Agent
- Uses tools
- Many steps
- Delivers outcome
Tools give Agents “hands and feet.”
The LLM supplies understanding and reasoning.
The Agent wires it together into finished work.
Try it for real
Try an Agent that searches, analyzes, and reports—not just chat.
Try Research Agent →