Systems Talks

Real conversations about the systems behind the work — why an agent never reviews its own diff, what a release gate actually verifies, and what breaks when two agents run in parallel.

Agent orchestration

Agent Systems

Shokunin Harness and the agent-orchestration work behind every Hard Machine Labs project.

EP. 01 · Agent SystemsHard Machine Labs · Systems Talk

The Shokunin Harness & Disposable AI Agents

Why the CTO agent never writes code, how born-clean subagents preserve context, and what a mechanical release gate actually verifies. A discussion around the architecture powering every Hard Machine Labs project.

Audio pending R2 publish

Agent SystemsHard Machine Labs · Systems Talk

How Shokunin Harness Kills AI Context Rot

How the Shokunin Harness keeps long agent sessions from degrading: born-clean subagents, disk-based state, and mechanical gates instead of one sprawling context window.

Audio pending R2 publish

EP. 02 · Agent SystemsHard Machine Labs · Systems Talk

Verifiable Business Execution with Shokunin Harness

The same mechanical gates, VERIFIED-never-DECLARED discipline, and CTO/Coder/Tester/Reviewer role separation that run the harness on code — applied to business execution outside of software engineering. Same rigor, a domain beyond the codebase.

Audio pending R2 publish

EP. 04 · MeteosureHard Machine Labs · Applied Systems

Meteosure et la preuve de tempête immédiate

Comment transformer la vérification météo pour les sinistres d'assurance d'un goulot d'étranglement de 3 semaines en une preuve déterministe sous 60 secondes. Un échange approfondi sur la rigueur du moteur Arago et la séparation stricte des rôles.

Audio pending R2 publish

In Production

Recordingindex-ai: Building the Machine-Readable Web Layer

PlannedAgent View: Making AI Retrieval Inspectable