Rank job applicants with one agent per person

A team used parallel Claude agents to score 115 candidates against a rubric and produce a ranked shortlist.

Done with Claude Code· Advanced

What they did
They exported their Notion hiring database into separate files, one for each applicant. A script then launched 115 Claude agents at once; each one read the scoring rules, checked the candidate’s GitHub and portfolio links, and sent back structured JSON scores. Another step sorted the results, and 10 extra agents acted as skeptics to challenge the leading candidates’ ratings.
How it went
The whole run finished in about 13 minutes and cost around $65, or roughly $0.57 per candidate. No applicant reached S or A, six were rated B, and the challenge step pulled the top two scores down from about 82 to about 75.
Tips
Parallel runs can get expensive because each agent may create its own cache. Using a cheaper model for scoring and reserving a stronger one for review can cut costs a lot.

Try it yourself with Claude Code

I have [number] applicants for [role] in [Notion database/spreadsheet]. Use this rubric: [paste rubric]. For each applicant, check their stored data plus public proof such as [GitHub/portfolio] against the rubric, and return one structured JSON score per person with brief evidence notes. Do not message anyone or edit any records. After the first ranking is ready, pause and ask me to approve any follow-up review step before you run any extra agents, then produce a ranked shortlist with short reasons and the total cost of the run.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on happycapy.ai ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.