{
 "text": "What running AI agents in production taught us 👇\n\n🔍 Week one: 611 of 612 agent sessions were empty polling cycles. About 31M tokens spent on “nothing happened”.\n⚡ The fix: event-driven agents. Idle token cost dropped by 95%.\n👀 Autonomy without visibility is just anxiety: leaderboard, work/idle accounting, daily reports.\n⏱ ~4 minutes for a board decision to become law on every agent.\n🔁 3 model migrations, zero harness loss.\n🛡 Money stays human-gated. Every action is an auditable event. Kill-switch per agent.\n\nEvery number here is a log, not an aspiration.\nMore at cil.by\n\n#AIagents #AI #Startups #Automation #BuildInPublic",
 "x": [
  "What running AI agents in production taught us 🧵\n\nWeek one: 611 of 612 agent sessions were empty polling. ~31M tokens on “nothing happened”.\n\nEvent-driven fixed it: idle token cost −95%.",
  "~4 min from a board decision to law on every agent.\n3 model swaps, zero harness loss.\nMoney stays human-gated; every action is an auditable event.\n\nEvery number is a log, not an aspiration → cil.by"
 ],
 "alt": [
  "Cover: What running AI agents in production taught us. Alex Shkor, Shkor Labs & Cooperative 42",
  "611/612 agent sessions in week one were empty polling cycles",
  "Make agents event-driven, not scheduled: idle token cost dropped by 95%",
  "Autonomy without visibility is just anxiety",
  "About 4 minutes for a board decision to become law on every agent",
  "Swap the model, keep the harness: 3 model migrations, zero harness loss",
  "Guardrails: revenue-weighted score, human-gated money, auditable events, per-agent kill-switch",
  "Every number here is a log, not an aspiration. cil.by"
 ]
}