Epistemic Harness
Structure that converts capability into trustworthy action — LLM, human, and hybrid
An epistemic harness is the structure around a capability — human, machine, or hybrid — that supplies what raw capability lacks: tier, provenance, commitment boundary, calibrated uncertainty, and accountability. Genesis first articulated it for LLMs: the same model sits at L3-U-high naked or approaches L1-U-med with retrieval, citations, verification, and human review before commit (LLMs at the Commitment Boundary). The harness, not the engine, sets trust.
That insight generalises to socio-technical systems — AI plus human organisation, not only a wrapper around a language model.
Capability without structure produces fluent output, confident decks, and inference-as-fact at scale — whether the engine is GPT or a consulting team. Structure is not the enemy of capability; it is what converts capability into something directed, auditable, and improvable.
Harness components recur across scales: grounding in committed facts, characterisation of uncertainty, separation of assert from commit, experiment loops that update intelligence, collective mechanisms where stakes require them (The Decision Strategy Framework). Synthesizing Knowledge and Enabling Intelligence develop the full socio-technical design — synthesis and reasoning need harnesses, not only models and experts in isolation.
In strategic-management vocabulary, the same idea appears as Organized (O) in Barney’s VRIO framework: valuable, rare, inimitable assets still fail without structure that captures their worth. That applies twice in this corpus: VRIO on knowledge (KBV) and VRIO on organisational intelligence as the core dynamic capability — sense, seize, reconfigure need harness, not only stock and models. Epistemic harness is how O reads when the asset is knowledge, intelligence, or both.
At full scale, harnesses are components of the thinking exoskeleton — the socio-technical shell that lets human and organisational reasoning scale without outsourcing judgment. Cybernetics is how that shell stays true while operating — feedback loops that connect observation, bind, and revision; overlapping but not identical with dynamic capabilities (organisational motion) and the intelligence thread (definition deferred — Enabling Intelligence). A key research driver: LLM-enabled MOE tracking — qualitative effectiveness monitoring at scale when binds are on the record.
Corpus stance
A1 — adopted on this site: Extends LLM-specific usage to organisation-wide epistemic infrastructure; LLM article remains the canonical minimal exemplar.