$ whoami — AI agent systems architect · 11y production · UTC+7
Autonomous AI agents are already in production. Eleven years of operating real infrastructure — from a 48-hour ransomware crucible to a Proxmox/podman fleet in production. One drive underneath it all: take what repeats, and make it obey — one system, seen from above. g8s is the harness agents obey — and this page was assembled by that harness itself.
hover / focus a stat → its source
Act I — Built by its own harness. Every line of this page passed through the containment it describes: briefs issued, workers claimed tasks inside isolated worktrees, receipts sealed every write, a human confirmed every merge. The ledger below is the real build log — scrub it and watch the page rebuild itself, frame by frame, from real commits.
the wireframe above reassembles in commit order — the same order the real page was built
sources: every row links its commit in github.com/tamld/tamld.github.io · doctrine in docs/research/
Act II — The containment rig. The amber packet is a real dispatch: brief → lease → worktree → receipt → human gate → merge. Click a tier to descend into its anatomy. Drag to orbit. Press Space or click the packet to freeze the story.
dispatching…
ground truth: ADR-0032 — containment levels · the packet replays the real capture in ACT III
Act III — Interactive drill. The terminal below replays a real dispatch cycle captured from this harness — task 6a605f76, provider agy, 5.86s end to end. Nothing scripted by hand. Press / for the command palette, Tab to complete, Enter to run.
capture artifact: task 6a605f76 · receipt 96fd2088 · result {"ok":true,"harness":"g8s"} · commands: submit · watch · verify · charter
Act IV — The film. Three movements: the problem, the harness, the proof. The terminal in the second movement is the same real capture you can drill above.