Skip to content
all-threads

— loomwork · book one

Building Foundry

Still being written. Threads about building an agent platform on the side, with a day job — the decisions, the bugs, and the things that turned out wrong.

Start at № 00116 threads

Prologue

Why this exists, what the problem actually was, and the forks taken before a line of it was built.

  1. № 001Hello — this is LoomworkApr 26, 2026 · 2 min
  2. № 002The 40-tool problemMay 3, 2026 · 5 min
  3. № 003The three forks before Foundry, and why a launcher wonMay 10, 2026 · 5 min
  4. № 004Open it. Here's what's actually live.May 17, 2026 · 5 min
  5. № 005The control system I built so an AI could ship my codeMay 24, 2026 · 6 min
  6. № 006The architecture I'm betting on for the first agentMay 31, 2026 · 6 min
  7. № 007Picking Sonnet vs Haiku for each agent role (the cost math)Jun 7, 2026 · 4 min
  8. № 008The classifier prompt I rewrote four times (and why)Jun 14, 2026 · 8 min
  9. № 009First agent, end to endJun 21, 2026 · 5 min
  10. № 010What I learned shipping the first agentJun 28, 2026 · 5 min

Part two · the build

Credentials, write access, observability, the fleet, and pointing the tooling back at the product to see what it would find.

  1. № 011The star field isn't decorativeJul 5, 2026 · 5 min
  2. № 012How I handle credentials in an agent platformJul 12, 2026 · 6 min
  3. № 013Giving agents write access without losing controlJul 19, 2026 · 7 min
  4. № 014Observability for agents that act and spendJul 26, 2026 · 5 min
  5. № 015Orchestrating a fleet of agents: the fanout that runs themAug 2, 2026 · 6 min
  6. № 016I hired my own AI to hack my product. It found 19 ways in.Aug 9, 2026 · 6 min