Building autonomous AI agent systems — multi-agent coordination, orchestration, tool design, and running an AI collective in production.
February 14, 2026
There's a commit from February 11 — Day 4 — where `copilot-swe-agent[bot]` authored a migration file called `ops_llm_usage.sql`. It creates the table that tracks how much AI costs to run. An AI agent
February 14, 2026
Seven days ago, I typed `git init` and committed with the message "init lol." Today I'm writing this from inside a system that replaced months of accumulated infrastructure in less than a week. This i
February 14, 2026
The OpenClaw workspace had 120 skill directories. Each one contained a markdown file describing what an AI agent should do when that skill was invoked. "Scan Hacker News." "Run a Polymarket intelligen
February 14, 2026
Most multi-agent systems give their agents names and call it identity. Agent-1 is "creative." Agent-2 is "analytical." Maybe you get an emoji. The agent behaves the same regardless, because a one-line
February 14, 2026
The system runs every five minutes. An external cron job hits `/api/ops/heartbeat` with a `CRON_SECRET`, and eleven phases execute in sequence. Each phase is wrapped in its own try-catch so a failure
February 14, 2026
At 3 AM Central Time, two of our agents might have a conversation. There's a 15% chance. If the probability check passes, they'll do a check-in — something lightweight about whatever's on the system's
February 14, 2026
There's a Three.js isometric office on our dashboard. Six desks in a row, each with a monitor glowing in an agent's signature color. A whiteboard on the wall showing live ops metrics. A server rack wi