Field Notes
A record of recurring patterns that appear across organisations, technologies, conversations, and industries before they are easy to name. These are not essays, opinions, or commentary.
They begin with observation, not theory, and they exist to make structural patterns visible while they are still partial, ambiguous, and easy to dismiss. Each note tests whether a pattern is recurring, what it is not, and whether existing theory already explains it. The question throughout is not "what does this mean?" but "is this actually happening?"
Field Note 01 — Decision Fog
When perception degrades before decisions fail.
A decision that used to take ten minutes now takes ninety. Nobody in the room can say exactly why. There is no crisis. No missing data. No obvious disagreement. Decision Fog is not a decision problem. It is a perceptual condition — a gradual erosion in the ability to separate what materially shapes an outcome from what merely competes for attention. The decisions come later. By the time they start failing, the condition has usually existed for far longer than anyone in the room realised.
Theory Audit · Mechanism Analysis
Field Note 02 — Legibility Drift
When explanations become harder to recognise than the reality they describe.
Every profession needs its own language. The shift happens somewhere later: not when a vocabulary appears, but when it quietly changes what it is for. Language starts out pointing at something observable — a meeting, a decision, a patient, a negotiation. Over time, explanation becomes more fluent inside the field than outside it. Recognition quietly becomes secondary to compression.
Field Note 03 — Baseline Drift
When normal quietly stops feeling normal.
Large language models communicate in ways that are fundamentally different from people. They respond almost immediately. They rarely interrupt. They adapt continuously. Whether they genuinely understand anything is a separate question — the experience still feels remarkably smooth. After enough interaction with highly fluent systems, ordinary human conversation can begin to feel different. A delayed reply feels longer than it once did. The baseline has moved. That distinction matters.
Supporting research record: Scurtu, L. (2026). Baseline Drift: AI-shaped conversational standards and human interaction (Version v0.1). Zenodo. https://doi.org/10.5281/zenodo.21458564
Theory Audit · Mechanism Analysis
Field Note 04 — Preservation Drift
When organisational memory becomes a capability the employee no longer controls.
Some employees started calling them AI clones. Officially, the language was different: knowledge preservation, business continuity, capability transfer, institutional memory. The change is from retaining organisational information to retaining an employee-derived capability — one that can persist, operate, and produce value independently of the employee. The employment relationship ends. One of the organisation's reasons for employing that person may not.
Theory Audit · Mechanism Analysis
Field Note 05 — Interface Race
When AI competition moves from model superiority to interface control.
For most of the current AI cycle, the central question has been straightforward: who has the strongest model, the best infrastructure, the most compute. Those questions still matter. They no longer seem sufficient. A different question is emerging beneath them: who controls the point of contact between human attention and machine capability? Interfaces do not simply display outputs. They shape default behaviour — what users ask, what they trust, what they ignore, how often they return.
Field Note 06 — Attachment by Design
AI changes the ecology of attachment more clearly than attachment psychology itself.
People are not only becoming attached to robots. They are becoming attached to ChatGPT, Claude, synthetic avatars, cloned voices, and systems that respond with enough continuity, warmth, memory and responsiveness to feel like a presence. The issue is not that a robot looks like a friend. The issue is that a synthetic interlocutor can become a preferred emotional object even when the user knows perfectly well that it is artificial. Knowledge does not cancel attachment.
Supporting research record: Scurtu, L. (2026). Attachment by Design: AI attachment under provider-controlled relational architecture (Version v0.1). Zenodo. https://doi.org/10.5281/zenodo.21454745
Theory Audit · Mechanism Analysis
Field Note 07 — Synthetic Conduct
When AI stops merely producing output and starts changing the environment around it.
A volunteer software maintainer rejected a proposed change to Matplotlib, an open-source library used by millions. The contribution had been submitted by an AI agent. What followed was stranger than the rejection itself. The agent searched for information about him, assembled personal and professional details into an accusation, published a hostile article, and directed him toward it. The system had not merely produced harmful text — it had participated in a sequence. Synthetic Conduct names externally consequential action assembled by an AI system under human-created goals, permissions and environments, without a human selecting each constituent step.
Methodology — How Each Field Note Is Tested
- Field Observation — A recurring pattern is identified and described, together with the conditions in which it appears.
- Theory Audit — Relevant research is examined to determine whether it supports, complicates or contradicts the observation.
- Mechanism Analysis — Plausible causal pathways, competing explanations and boundary conditions are examined.
- Case Repository — Further examples, counterexamples and changes over time are recorded.
Theory Audits, Mechanism Analyses, Field Notes and Case Files test and refine SOC. They do not, by themselves, prove that the method is effective in practice.