How our recruiting runs on agents

Most technical recruiting comes down to research and judgment: finding the right people, and deciding whether they clear the bar, which takes taste. Since most of the best engineers in the world don't apply to jobs, they have to be reached out to. We use Claude managed agents to automate the repetitive parts of that proces which is going through large swaths of data, and the GTM motion of reaching out to candidates.

The architecture

Two layers do most of the work.

Superserve — the layer that sits between our team and the model. It holds our tools (GitHub, public profile enrichment, our candidate trackers), manages the prompts and search specs for each role, and exposes the agent through the interfaces our team actually uses: a CLI to start, then Slack.

Anthropic's cloud agents — where the actual reasoning happens. We hand the agent a goal ("find ~15 high-signal engineers who maintain this project, public sources only, with evidence") and it runs the loop: clarify the target, propose a plan, execute against the tools, and produce structured output. Running it as a managed agent means we don't babysit infrastructure — we describe the work and the runtime handles the steps.

How we built it

1. It started as a CLI: finding research scientists with semantic search

The first version was built with Claude running locally. We gave it access to public APIs for finding people — GitHub and a few other third-party sources — and queried candidates against public datasets, grading them on our own criteria. It was a solid proof of concept.

2. We connected it to Slack and to our internal app

Next, we realized we'd have to deploy a search like this for every new role we took on. Since we already talk to many of our customers on Slack, that's where we connected it. Now a candidate's grade can be reviewed by a customer too, and their feedback flows back into our database as we reach out to them.

3. Next: customers run it themselves through our web UI

The version we're building now opens this up to our customers directly, through our web app — which also includes an ATS. Instead of a search being something we run on their behalf, they'll describe the role in our web-based UI and be provided an end-to-end agentic recruitment service.

The future will run on agents

This workflow lets our recruiters be far more efficient with their time and serve customers faster than the incumbents can. The judgment, the taste, the relationships, the read on whether someone will actually thrive on a two-person founding team — those all still require human skill. But because of a workflow like this, we're able to service many more clients simultaneously with a lean team.