The systems-engineering curriculum on building deterministic runtime substrates, sandboxes, verification sensors, and governance layers around foundation models.
Autonomous coding is defined by a simple systems equation: Agent = Model + Harness. The model supplies probabilistic reasoning; the harness provides the deterministic control plane that makes execution reliable, bounded, and reproducible.
Design repositories that prevent autonomous agents from creating architectural decay and unmaintainable code.
Implement hermetic sub-second execution sandboxes, permissions, and deterministic verification sensors.
Understand the business ROI, unit economics, empirical benchmarks, and real-world case studies.
The origin story (Hashimoto, Lopopolo), the Intelligence Fallacy, disciplinary boundaries, and the H0–H3 maturity ladder.
Ashby's Law of Requisite Variety, the Hashimoto Loop (Mistake → Permanent Control), and Tessl Inner/Outer/Meta loops.
Paired contrast designs, 6-axis Pareto profiles, Harness-Bench, Doc2Harness, Terminal-Bench, and Thoughtworks ROI analysis.
The formal H tuple, 11 component responsibilities (Zhong & Zhu), failure modes prevented table, and H0–H3 technical specifications.
Attention dispersion beyond 64k tokens, ACE, ACM (5 primitives), AdaCoM, Tree-sitter AST maps, and progressive disclosure.
Threat modeling, Docker vs. gVisor vs. Firecracker, non-interactive supervisors (PAGER=cat), and Copy-on-Write git worktrees.
Princeton SWE-agent Agent-Computer Interfaces (ACI), sub-5s Test Impact Analysis (TIA), and Agentless sequential pipelines.
HERMES Dev-Primitives, ArchUnit boundaries, cyclomatic complexity limits, and Model Context Protocol (MCP) tool standard.
Mitigating dead code and entropy, background asynchronous GC subagents, Knip dead-code elimination, and doc sync.
The primary security boundary axiom, 6 lifecycle risk phases, Allow/Ask/Deny matrix, multi-agent fleet risks, and episode packages.
OpenAI Codex 1M LOC experiment (Ryan Lopopolo), Hashimoto adoption protocol, and Thoughtworks cybernetic governance.
Complete runnable Python ETCLOVG implementation and source dissections: Claude Code, Codex App Server, OpenHands, LangGraph.
7 major production incident post-mortems mapped to permanent controls, cross-domain effect sizes, and the Systems Authority Blueprint.
The Scaffolding Treadmill, Harness-Zero, HAD, Co-Harness, RLVR verifiers (DeepSeek-R1, o3), and the Invariant Boundary.
We haven't launched yet. Leave your email to receive technical briefings and open-source harness releases.