Step 2 — Set Up Claude API

python import anthropic client = anthropic.Anthropic()

Step 3 — Define Tools

python tools = [ {"name": "searchkb", "description": "Search knowledge base", "inputschema": {...}}, {"name": "sendreply", "description": "Reply to email", "inputschema": {...}}, {"name": "escalate", "description": "Hand off to human", "inputschema": {...}}, ]

Step 4 — Agent Loop

python def runagent(email): messages = [{"role": "user", "content": email.body}] while True: resp = client.messages.create( model="claude-sonnet-4-6", maxtokens=1024, tools=tools, messages=messages ) if resp.stopreason == "endturn": return resp.content[0].text for block in resp.content: if block.type == "tooluse": result = executetool(block.name, block.input) messages.append({"role": "assistant", "content": resp.content}) messages.append({"role": "user", "content": [{"type": "toolresult", "tooluseid": block.id, "content": result}]})

Step 5 — System Prompt

You are a customer support agent for [Company]. You read incoming emails, search the knowledge base, and reply if confident. Escalate to human if: - Refund or billing issue - Customer threatens legal action - Knowledge base has no relevant article

Vercel Edge Function, AWS Lambda, or your own server. Trigger on new email via webhook.

No loop limit: agent could loop forever — set max iterations - No human-in-loop: critical actions need approval - No logging: you need observability to debug - No cost cap: token usage adds up fast

For automation orchestration, read [how to use n8n](https://vibecodingturkey.com/en/rehberler/how-to-use-n8n).

How to Build an AI Agent 2026 — Guide

Q: Step 1 — Pick a Use Case

Concrete example: "Customer support agent that triages incoming emails." Tools needed: - Read email - Search knowledge base - Send reply - Escalate to human

How to Build an AI Agent 2026

AI agents are LLMs with the ability to call tools, observe results, and plan multi-step workflows. This guide builds a real agent in 60 minutes. What Makes an Agent A bare LLM answers your question. An agent: 1. Plans steps 2. Calls tools APIs, DB queries, web search 3. Reads results 4. Decides next action 5. Repeats until done Step 1 — Pick a Use Case Concrete example: "Customer support agent th…

All guides

Related guides

How to Start AI SaaS 2026
How to Use v0 by Vercel 2026 — Complete Walkthrough
How to Become a Vibe Coder: 2026 Roadmap
How to Build a Chatbot 2026
How to Build Supabase Edge Functions 2026