An AI agent is a system where AI reads input, decides what to do, takes actions, and returns results — autonomously. Examples: an email triage agent, a research bot, a customer support agent, a competitive intel scraper. This post shows you how to build your first one in 1 day, using n8n + Claude or ChatGPT API.
Step 1 — Pick a real, painful task to automate
Don't build a generic 'AI assistant'. Pick one specific painful task in your life: 'triage my work email and draft replies', 'summarise my LinkedIn DMs daily', 'monitor 5 competitor websites for changes'. The narrower, the better.
Step 2 — Set up n8n (free, takes 15 min)
- Go to n8n.io — sign up for free cloud or self-host
- Create a new workflow
- Add a trigger node (schedule, webhook, or email)
Step 3 — Add the AI brain
Add an HTTP Request node calling Claude or ChatGPT API. Pass it the input from your trigger and a clear system prompt: 'You are an email triage agent. Categorize this email as URGENT, IMPORTANT, INFO, or SPAM. If URGENT, draft a 3-sentence reply.' Get an API key, paste it in, test.
Step 4 — Add actions
- Send the AI's output to Slack / WhatsApp / email
- Save categorised emails to a Notion / Google Sheets database
- Trigger downstream workflows for URGENT items
Step 5 — Test, iterate, deploy
Run the workflow on 10 sample inputs. Tune the prompt until 9/10 outputs are useful. Schedule it to run hourly / daily. Done — you have a deployed AI agent.
Common mistakes building your first agent
- Too broad scope — narrow down to ONE specific task
- No retry logic — APIs fail; add retries
- No human-in-the-loop for high-stakes decisions
- Forgetting cost monitoring — Claude/ChatGPT API can rack up fees on a runaway loop