Research agents as a scouting team

what I built

Repeatable research runs: complaint scans across social sites, broad market scans with agents working in parallel, and a daily watch of new AI releases.

Claude Sonnet agents · last30days · web search

what happened

• One deep research run used 104 agents to check 25 claims about a sales channel. 12 were confirmed and 13 refuted.
• A complaint scan with 2 agents used about 165 tool calls and found 3 new small ideas.
• A broad scan with 4 parallel agents used about 300 tool calls.
• A market check rejected 16 of 23 claims about gaps in one market.
• The first daily release watch judged 3 ideas and found 0 worth building.

what broke

• One social site blocked access after 2 items, and the search budget ran out mid run.
• The daily watch could not be scheduled, because creating the task was blocked by a permission check.

decision

None of the ideas was built.

• Ideas from headline AI releases are already being sold by funded startups within days.