Add Many Hands Engineering to Foundations

[mseeks]

Disclosure first: this is a self-submission. Please take it or leave it on its own merits.

Of course AI helped write the text along with my use of it in heavy cross-examination & review, but the underlying research was done by me (a human 😉) and all inferences, cross-pollination of ideas, and extrapolations are my own.

The piece is called Many Hands Engineering. It's a working handbook in long-form free PDF. It tries to articulate a vocabulary and framework for the layer that sits above the per-agent harness. Its central moves: a planned/emergent spectrum used as a placement discipline; signals and a governed commons as the substrate multiple agents coordinate through; pace layers (governance, architecture, capabilities, orchestration, agents, invocations); and the steward as a lower-frequency, higher-authority human role upstream and downstream of agent execution.

§4.2 of the handbook is dedicated to harness engineering specifically and traces the lineage — Hashimoto's originating definition, the OpenAI Codex piece, Böckeler's analysis, HumanLayer, Anthropic's long-running harness work, Stripe's eval harness. The framework explicitly positions harness engineering as the per-agent layer of "terrain" it builds on, not an alternative to it. So the relationship to the Foundations section is direct.

The handbook also flags claims by epistemic layer — evidence-backed observations vs. working design principles vs. organizing metaphors — rather than presenting itself as settled methodology. The grounding pulls from distributed systems (CAP, FLP, Ashby), commons governance (Ostrom), pace layers (Brand), Conway, and the cybernetics tradition. It's a principled hypothesis, not a proven framework.

Thanks for maintaining the list either way!