Models
Each conceptual model names a pattern — something that keeps appearing across different organizations, roles, and contexts when AI enters a workflow. The models don't prescribe solutions. They help you see what's actually happening.
Plausibility Anchor
Once an answer appears reasonable, people stop looking for the same kinds of mistakes. Verification continues, but it no longer starts from neutrality. The anchor forms before verification finishes, which is exactly when missing information becomes hardest to notice.
Trust-Verification Decoupling
Usability improvements can remove uncertainty cues faster than they remove uncertainty itself. As AI outputs become cleaner and more coherent, the signals that once triggered verification behavior disappear — before the underlying uncertainty those signals were indicating has been resolved.
Coherence vs Correctness
Coherent, well-structured work used to require genuine understanding to produce. That assumption quietly shaped how expertise, correctness, and quality were evaluated. When coherence becomes cheap to generate, the signal remains — but the properties it once indicated no longer reliably follow.
Machine-Consumed Infrastructure
Documentation, APIs, and knowledge systems were built for humans — readers who interpret, infer, and exercise judgment. When AI agents become primary consumers, they consume literally, without the ambient context human readers supply. The infrastructure isn't broken. It's being used by a consumer it wasn't designed for.
Expertise Relocation
When production costs drop, expertise doesn't disappear — it relocates from production to evaluation. The expert who previously did the work is now the person who determines whether the work is acceptable. Organizations that measure expertise through production miss this shift entirely.
The Proxy Was the System
Institutions substitute observable signals for conditions they can't directly observe — a resume for capability, a review score for future research quality. The substitution held because producing the signal required the underlying condition. AI breaks that coupling, making the signal cheap to produce without making the condition any easier to see.
Premature Operational Confidence
AI-generated output can look ready almost immediately, because it reads clean and coherent. That's never been the same thing as being verified — the two used to move together, since confirming correctness took the same effort that made work feel ready. When one threshold can be crossed without the other, work proceeds on the strength of a check that was never actually made.
Human-in-the-Loop Is Not a Strategy
Organizations treat a human in the workflow as though presence alone creates oversight. It doesn't. Oversight is a capacity — time and expertise together — and it can be exceeded long after the checkpoint itself keeps running. The approval still happens. The sign-off still gets logged. What used to happen inside that box can quietly stop.
AI Rewards Confidence Before Correctness
Organizations often behave as though rewarding speed and rewarding correctness are the same thing. They aren't — the signal that determines recognition arrives before the signal that reveals whether the work was actually right. When no correction signal ever arrives, rewarded confidence doesn't just go uncorrected. It becomes precedent.
Coming Soon
Approved frameworks currently being developed for publication.
- Validation Is Invisible Work — Why the work that prevents failure is often invisible to the systems used to measure contribution.
- The Integrator Becomes the Bottleneck — As production becomes easier to scale, the scarce work shifts to maintaining relationships between outputs.
- Exposure Is Not Incorporation — Why exposure to information and actual behavioral adoption are often mistaken for the same thing.