Count them, because the vendor's number does not. Nathan Drezner's October 28, 2026 LangChain post says most proactive agents are cron tasks and more than 70% of Managed Deep Agent runs on LangSmith come from a schedule rather than a person; that counts runs nobody typed. When I audited my fleet of 39 appointed agents, 13 of 39 were orphans, running reliably and producing clean output that nothing outside the fleet read, several for over a month, and 13 of 39 had drifted to the same 7 to 9 a.m. window on their own; the fleet is 44 today and 15 still sit there. A cron server can jitter its load, and it cannot jitter one reader's morning. Sydney Runkle's four human touchpoints put a person before the sensitive call and before the output ships, and never name who opens it. So for every scheduled run, write the name of the person who opens it, count the blanks, and publish that share next to the vendor's. The check is not the agent's log: count what outside the fleet consumed the output, a page, a ritual, another agent, a decision. The heartbeat says it ran. Only something downstream says it mattered.
Count them. The vendor's 70% counts runs nobody typed; the number next to it is runs with a named reader.
The three posts, the fleet counts, and the check are in 70% of LangChain's Agent Runs Are Scheduled. Count the Ones Somebody Read.. The three lines on the charter are in 44 Appointed Agents, One Attention Budget: Three Lines on the Charter, and the April audit is 39 PM AI Agents Deployed: What Stuck, What Died, and Why.
Last reviewed 2026-10-30 · 1 min read