The Structural Dynamics of AI Adoption

AI capability does not compound across the organisation.

Individual interactions do not become shared institutional memory. When employees invest cognitive effort into AI systems, that investment does not accumulate anywhere retrievable. Skill remains personal. When the individual leaves, the capability leaves.

Core governance risk: Capability concentration and key-person dependency.

First named by Richard Foster-Fletcher in 'One Player Game,' What Still Matters, 2026.

Work bends toward the model's default patterns.

AI systems produce fluent, competent, undifferentiated output. Preserving distinctiveness requires additional effort. Over time the path of least resistance wins. The organisation drifts toward competent sameness without a decision to do so.

Core governance risk: Competitive differentiation erosion.

First named by Richard Foster-Fletcher in 'The Gravity of the Generic,' What Still Matters, 2026.

When AI drafts first, its framing and causal structure persist even after human edits.

Most revision changes wording while leaving the underlying argument intact. The structure AI created remains. The assumptions AI embedded survive. Framing decisions originate with the model, while accountability attaches to the human editor.

Core governance risk: Decision provenance failure.

First named by Richard Foster-Fletcher in 'Residual Logic,' What Still Matters, 2026.

The system presents one plausible path as the answer.

Equally viable alternatives are not rejected; they are never surfaced. The model's ranking function, summarisation logic, or selection criteria eliminate options before decision-makers arrive. The presented option may be defensible. The absent options may have been equally defensible.

Core governance risk: Strategic option truncation.

First named by Richard Foster-Fletcher in 'The Single-Path Illusion,' What Still Matters, 2026.

As judgement is routinely offloaded, human capacity to hold uncertainty, contest framing, and reconstruct reasoning weakens through disuse.

When the AI fails or produces something subtly wrong, the human is less equipped to notice, less able to diagnose the problem, and less capable of reconstructing the work from first principles.

Core governance risk: Capability atrophy through dependency.

First named by Richard Foster-Fletcher in 'Brittlement,' Structural Dynamics of AI Adoption, 2026.

The underlying system changes in ways the organisation cannot fully observe.

Providers update weights, adjust filters, and modify behaviour without notice. The same prompt can yield different output over time. Yesterday's evidence becomes a time-stamped artefact, not a repeatable result. Reproducibility is structurally unavailable.

Core governance risk: Reproducibility failure and evidence degradation.

First named by Richard Foster-Fletcher in 'Model Volatility,' Structural Dynamics of AI Adoption, 2026.

Liability for AI output falls on the individual mandated to use the system, not on those who imposed or built it.

The professional is mandated to use an AI system they did not choose and cannot opt out of. When the governed output fails, the liability falls on the person who signed it, while the organisation that mandated the tool and the vendor that built it remain untouched. The same governance that prevents the wrong answer produces the weak one, and the person carries the cost either way.

Core governance risk: Accountability transfer and misattributed liability.

First named by Richard Foster-Fletcher in 'Blamefall,' Structural Dynamics of AI Adoption, 2026.