An organisation makes Commitments from what it believes about the world. It acts on those beliefs, encounters consequences and has an opportunity to revise its understanding.
That opportunity depends on whether the consequences can be examined openly.
If evidence becomes difficult to question, the organisation’s Commitments become harder to judge. If unwelcome outcomes become difficult to discuss, the organisation loses the means to improve them. Work continues. Decisions are made. The shared understanding behind those decisions can drift further from reality.
Over the last two years, this relationship between judgement, Commitment and Organisational Learning has shaped my thinking about Reflective Labs. In The Physics of Organizations, I described the adaptive cycle through which an organisation brings knowledge together, commits to action and revises its understanding from what happens.
AI changes the cost of running that cycle. It also changes how easily we can tell whether parts of it have happened.
Judgement and Cognitive Surrender
Consider a proposal arriving for approval. It contains a clear recommendation, a financial justification and a delivery plan. The document looks ready for a decision.
The reviewer still needs to establish whether the recommendation fits the situation. Someone must understand which assumptions carry the argument, whether the proposed work can be delivered and what the organisation would become responsible for if it proceeds.
Generating the proposal and establishing those conditions are different kinds of work. AI can accelerate the first while leaving the second unresolved.
That difference becomes consequential when the appearance of a finished document is taken as evidence of a finished judgement.
In their 2026 research, Steven Shaw and Gideon Nave at the Wharton School, University of Pennsylvania, describe one way that happens: cognitive surrender.
Their account extends Daniel Kahneman’s distinction between fast intuition and deliberate reasoning with System 3, external artificial cognition. Cognitive surrender occurs when a person accepts the system’s answer with little scrutiny and allows it to determine the judgement.
Their research involved three experiments and nearly 10,000 trials. In the first experiment, accurate AI advice raised performance by 25 percentage points above the baseline of people working without AI. Faulty advice lowered it by 15 percentage points below that baseline. Confidence increased even when the advice was wrong. These were controlled reasoning tasks, with implications that need to be examined in other settings. Wharton’s account of the research describes both the findings and interventions that helped participants question faulty advice.
The organisational concern follows from the separation between confidence and reliability. A person can feel better equipped to approve a proposal while having a weaker basis for doing so.
The approval then carries that judgement into other people’s work.
The Capability Mirage and the Approval Record
In September, Melissa Swift of Anthrome Insight and Teryluz Andreu and Dolores Hernandez of Axialent described a related difficulty in MIT Sloan Management Review. Drawing on interviews with practitioners and organisational specialists, they called it the capability mirage. Generative tools can improve the appearance of work while obscuring the knowledge and judgement behind it. Managers and colleagues find it harder to assess who understands a problem and where support is needed. Their article examines the consequences for evaluation and trust.
A polished proposal used to provide a rough indication that someone had worked through its contents. It was always an imperfect signal. AI makes that signal less dependable.
The proposal may be sound. It may also contain assumptions that nobody involved has examined. Both can arrive in the same format, with the same apparent completeness.
In No room for slop, I described slop as a structural condition: fluency that exceeds the available trace of the judgement behind it.
That definition directs attention to what the organisation can establish about a proposal. What evidence supports it? Which assumptions were challenged? What changed during review? Who accepts responsibility for proceeding?
A timestamp and an approval flag answer very little of this.
Neither does the time spent reviewing. Three hours can contain careful examination or repeated distraction. Thirty seconds can be sufficient for a familiar, bounded decision. Duration alone tells us little about the quality of judgement.
What matters is whether the basis for the Commitment is clear enough for others to act on it and revisit it when circumstances change.
Defensive Organizing and Unwelcome Evidence
The difficulty grows when an organisation has already declared its AI programme a success.
People may be producing drafts faster while review queues grow and delivery remains slow. Those observations can both be accurate. Dependencies, verification and rework may explain the gap.
The organisation needs to investigate that gap. Its ability to do so depends partly on how leaders respond to the people who describe it.
In September, Keith Cotterill drew my attention to three recent MIT Sloan Management Review articles that helped me think about this relationship. Alongside the capability mirage was work by Declan Fitzsimons, Gianpiero Petriglieri and Jennifer Petriglieri at INSEAD.
Their September article drew on a four-year study of a professional services firm, first reported in an academic paper in 2025. They examined how a transformation could fail despite committed leaders and a plausible strategy.
Their explanation centres on defensive organizing. Anxiety becomes difficult to acknowledge, so organisational activity begins to serve the need for reassurance. A vision offers comfort. Teams reporting difficulties become convenient targets. Meetings and communication campaigns continue while confidence in the transformation weakens. The researchers describe a progression through a comforting vision, a convenient foe, empty ritual and eventual reckoning. Their account concerns the relationships through which people handle uncertainty.
Read alongside the capability mirage, this suggests a reinforcing pattern.
Polished output can reassure leaders that progress is being made. Leaders who depend on that reassurance may become less willing to examine the weaknesses in the output. People learn which observations are welcome. The evidence reaching the next decision becomes less representative of what is happening.
This connection is my interpretation of the studies. It is a possibility organisations should be able to recognise and investigate.
It also connects directly to the adaptive cycle. Organisational Learning depends on being able to compare an expected outcome with an observed one. When that comparison threatens the accepted story, the organisation may preserve the story by restricting the evidence.
The next Commitment then inherits the same weakness.
Otto Scharmer of MIT Sloan approached the problem from another direction in July. In “Leadership’s Blind Spot in the Age of AI”, he argued that organisations risk an intelligence monoculture when investment in AI crowds out the human capacities through which people interpret situations together. His framework distinguishes machine intelligence from embodied, relational judgement and the collective capacity to sense emerging possibilities.
For me, the practical question is how an organisation maintains the capacity to notice that its current understanding is inadequate.
That capacity needs attention, relationships and people who can bring uncomfortable observations into a decision. It also needs somewhere for those observations to remain available after the meeting ends.
Leadership Responsibility and Decision Records
This is where the research has changed what I ask of the systems we are building.
Fitzsimons and the Petriglieris place responsibility firmly with leaders. They describe the courage to acknowledge anxiety, the curiosity to hear challenging perspectives and the care to recognise people’s exhaustion. A workflow cannot establish those conditions by itself.
Software can preserve an objection. Leaders determine whether raising it carries a penalty.
Software can connect a decision to its evidence. People determine whether contrary evidence receives a fair hearing.
Both responsibilities matter. A decision record is useful when the organisation is willing to act on what it reveals.
System 2b, Commitment and Organisational Learning
In my earlier writing about System 3, I called the practice of deliberate engagement with AI System 2b. Writing down intent before generating a proposal is one useful part of that practice. Intent itself can also develop through discussion with an AI system.
The requirement is that human judgement remains active throughout the work.
People need to understand what they are trying to achieve, examine the evidence and decide what uncertainty they are prepared to accept. They need to be able to challenge the proposed action and see how that challenge affects the decision.
A Commitment is an authorised undertaking on which others can rely. It binds the organisation to a specific outcome under stated conditions. Resources are allocated and responsibility for carrying it out is clear.
That makes the conditions for closing a review consequential.
A review may approve a proposal, return it for revision or decide to stop. Its outcome must be clear. Approval requires the relevant authority, evidence supporting the proposed Commitment and explicit treatment of material objections. The people expected to act must understand what they are responsible for and what would require reconsideration.
These conditions provide a stopping point for deliberation within the agreed scope. New evidence can reopen it. Closing a review leaves the outcome of the work still to be observed.
Organisational Learning continues through execution and review of the outcome. Experience must affect the understanding used in subsequent decisions. The final essay returns to what must be true for that loop to close.
We are designing Organism.zone as a workspace to support that continuity: from the question being examined through the Commitment being made to the outcome that tests it. The reasoning should remain available as people move between meetings, responsibilities and applications.
The standard for the system follows from the work it is meant to support. It should help people establish what they are committing to, understand why the Commitment is justified and recognise when it needs to change. It should help the next decision benefit from what happened after the previous one.
An organisation’s ability to learn depends on what happens when experience challenges its expectations.
AI makes that condition more consequential. It allows proposals and decisions to multiply quickly. The organisation still has to establish which of them it can stand behind and what it has learned from carrying them out.
The next essay examines why faster individual work can leave delivery almost unchanged and where to look for the missing connection.
Core thesis: An organisation’s ability to learn depends on whether the evidence behind its Commitments and the outcomes they produce can be examined openly.
The idea you can't unsee: A polished proposal and an approval flag show that a document passed through a process, while leaving the quality of judgement unrecorded.
Vocabulary shift: "approval recorded" → "judgement made accountable"
Connects to: Article 2 (When Individual Speed Outruns Shared Understanding), Article 1 of The Physics of Organizations (The Organization as an Adaptive System), Report 11 (No room for slop)
Version: 1.1 / 2026-11-07