I Went to the Pool. The Swarm Did the Work.
Rule-based shipped fast. AI found what the rules missed — and doubled the at-risk bucket.
April 3, 2026 · Build log
Left Claude Code running and went to the pool. That's the sentence that describes most of April 3.
Before I left, I handed off a clean task: find the updated Blueprint Swarm skill, run the full pipeline, publish the playbooks. No subprocess calls, no Anthropic API — just the Agent tool, the way the swarm is designed to run. The orchestrator picked it up from there.
First thing it did: pull fresh revenue data from the CRM. Prep the full active account base across classification batches. Then run a rule-based pass first to get the playbooks published fast — three playbooks shipped to playbooks.blueprintgtm.com before the AI agents even finished their first wave. That's the lazy phase. Get something out the door while the better work runs in parallel.
The better work: AI agents on the same batches. The comparison was instructive. Rule-based put about a third of the at-risk accounts in the "Engage or Lose" bucket — accounts showing silent disengagement signals. The AI put roughly twice as many there. The rules were miscategorizing those accounts as product issues or payment problems because they pattern-matched on keywords. The AI read the full context and recognized the difference between a keyword and a root cause. The "Engage or Lose" bucket doubled in size. The playbooks got republished with the corrected distribution.
Then the swarm stayed running through the night — hundreds of batches of churned account forensics, five accounts per batch. Peak throughput hit sub-4 minutes per batch. One consistent pattern across the entire run: CRM labels were symptom labels, not root-cause labels. An account tagged "Charges and Fees" turned out to be a bankruptcy. An account tagged "Product" was a sales misrepresentation. An account tagged "Unresponsive" was a ghost that had never activated. The AI flagged the mismatch every time; the rules would have just inherited the bad label.
Next: run the same pass over active accounts and see if the pattern holds.
What Annual Adds
This is what I built today. Annual subscribers run the same tools.
Every tool I ship. Edge Copilot installs to your Claude Code — talk to all my knowledge, every method, every data source. Current: Edge Copilot, AutoClaygent, Agent 7, Who to Target and What to Say, Blueprint Cloud. Whatever ships next is included.
All 3 courses: Who to Target and What to Say, Agent 7, AutoClaygent.
Weekly office hours.
License key hits your email after you upgrade.