Your First AI Hiring Agent

SETUP STEPS, CONFIGURATION, TROUBLESHOOTING

The full walkthrough, start to finish. Set aside an afternoon and follow it in order.

STEP 1: CONNECT THE 3 TOOLS

1. Create a free Claude account at claude.ai and a free n8n cloud account at n8n.io (or self-host if you're comfortable with that).
2. Create a new Google Sheet with columns: Name, LinkedIn URL, Current Role, Current Company, Match Score, Outreach Status, Date Added.
3. In n8n, add a Google Sheets credential and connect it to the sheet you just created. This is the only credential you need for week one.
4. Add a Claude (Anthropic) credential in n8n using an API key from console.anthropic.com. This is the only paid piece, and costs are usually £10-30/month at this volume.

STEP 2: WRITE YOUR FIRST BRIEF

Before you build the workflow, write the brief the agent will screen against. Use the same format as the JD Writer prompt: role title, 5-8 responsibilities, what success looks like at 90 days, and 2-3 non-negotiables. Paste this brief into a note you'll reference in the next step. A vague brief produces a vague shortlist, no matter how good the automation is.

STEP 3: BUILD THE FIRST WORKFLOW

1. In n8n, create a new workflow with a Schedule Trigger set to run once daily.
2. Add an HTTP Request or scraping node that pulls candidate profiles matching your search criteria (a Boolean string from the Boolean Builder prompt works well here).
3. Add a Claude node that takes each profile and your brief, and scores it 1-10 with a short reasoning line, using a prompt similar to the CV Screener agent.
4. Add a filter node that only keeps profiles scoring 7 or above.
5. Add a Google Sheets node that appends the filtered results as new rows.
6. Add a Telegram or Slack node (optional) that sends you a daily summary count.

STEP 4: RUN AND REVIEW

Trigger the workflow manually the first time and watch it run end to end. Check the sheet. Are the scores sensible? Are the reasoning lines specific to your brief, or generic? If generic, tighten the brief and re-run. This iteration loop, not the initial setup, is where most of the quality improvement happens.

COMMON TROUBLESHOOTING

• Empty results: your Boolean string is too narrow. Loosen one AND condition and re-test.
• Scores all cluster around the same number: your brief's non-negotiables are too vague for Claude to differentiate against. Make them more specific and measurable.
• Workflow times out on large batches: add a batching node that processes profiles in groups of 20-50 rather than all at once.
• Sheet duplicates: add a lookup step that checks LinkedIn URL against existing rows before appending.

THE FIRST WEEK

Expect the first 2-3 days to be tuning, not results. By day 4-5 the scoring should stabilise and the shortlist quality should be consistent enough that you're spending 10 minutes reviewing instead of 3 hours searching. If it isn't, the brief is still the most likely culprit, not the automation.