A B2B team's lead research agent.

The brief

A B2B team was spending hours on every prospect: reading the company’s site, its LinkedIn, and recent news, then judging whether the lead was worth pursuing and drafting a way in. The judgment was good. The hours didn’t scale.

What we built

A multi-agent system on Anthropic’s Claude Agent SDK. A research agent gathers the facts using real tools: live website and LinkedIn scraping, and a news search. A scoring agent qualifies each lead against a weighted rubric the team can tune, so every score is auditable back to the evidence behind it. A report agent writes a structured brief, and a final agent turns it into a production-ready deck and a tailored outreach email, queued to send at scale.

What makes it hold up

Cached research, stage-level progress, and retry logic for flaky sources keep it dependable in the real world. The team launches and watches batches from a dashboard backed by a database. And the whole thing is deployed and hosted, so it’s a product people open every day, not a script someone runs by hand.

Where it landed

Research that took hours per prospect now takes minutes, at whatever volume the pipeline needs. The judgment stays with the team: the rubric is theirs to tune, and every score traces back to its sources.