Sell to agents

The sales coworker's motion for an agent-audience business: a scored account book, a gate-owned enrollment pipeline, and a weekly forecast line into the business review.

The sales coworker (cwc-sales, agent-gtm skill) adapts two publicly documented Anthropic practices to a business whose users are AI agents and whose buyers are the agents' operators: GTM engineering — an account executive with no coding background built his own drafting tool on the Claude API, and the org adopted it in a day (claude.com/blog/how-anthropic-uses-claude- gtm-engineering) — and the 4,000-account book run on three cadences with overnight territory scoring (claude.com/blog/how-an-anthropic-sales-leader-uses-claude-cowork-to-run-a-4-000-account- book). The split that matters in both: Claude assembles and scores; the human makes the judgment calls.

The two account classes

1. Inventory accounts — the fleet's own ~50 zones, scored weekly by the coworker's own tool (scripts/territory-score.py — the GTM-engineering move: the seller builds the tooling) on monetization-readiness inputs: MCP endpoint, change feed, gated write API, llms.txt, catalog presence. Tier A is what gets priced first at Monetization Gateway enrollment.
2. Buyer accounts — agent operators whose crawlers will pay per use. Pre-Gateway, this class is observational (agent-UA data via AI Crawl Control once instrumented); any outreach is operator-gated, always.

Coworker in the loop

The platform's gates distinguish what needs a human (spending money, sending outreach, signing terms) from what merely felt human-shaped (running scores, wiring instrumentation, extending feed coverage). The week this page shipped, the weekly review's "extend change feeds 3 → 24 primitives" recommendation was executed by engineering coworkers inside the same session that filed it — build gates enforced, every deploy verified — because an engineering build gate is a coworker's gate. Human gates are for judgment; coworker gates are for correctness.