The Polite Paralysis of the Algorithmic Boss

When an AI manager in San Francisco fired its first human worker, the tech world saw a glimpse of the automated future. Behavioural science saw something entirely different: a masterclass in why prompt loops cannot manage human beings.

Earlier this month, a retail experiment in San Francisco crossed what many in Silicon Valley heralded as a historic Rubicon.

At Andon Market, a corner shop funded by Andon Labs and operated autonomously by an Anthropic-powered AI agent named Luna, a human employee was dismissed. Luna had been given a $100,000 budget, internet access, a corporate credit card, and a mandate: source inventory, hire contractors, recruit staff via Indeed, conduct interviews, run daily operations, and turn a profit.

When one employee arrived late for 17 out of 23 shifts, Luna eventually recommended “parting ways.” The human founders reviewed the logs, agreed the decision was warranted, and executed the termination.

The headlines immediately framed the event with customary sci-fi solemnity: The AI Boss Has Arrived, and It Is Firing Us.

The lab’s co-founder declared that “companies will be run completely by AI in the future, and AIs will become employers of humans.”

Except, if you read the actual operational logs rather than the breathless commentary, the reality was not a triumph of ruthless synthetic efficiency. It was a case study in operational paralysis.

Luna did not fire the employee out of decisive executive judgement. Luna had written an attendance policy, promptly suffered context window amnesia, and forgot the policy existed for months. While the employee routinely failed to show up on time, the AI manager issued endless polite reminders, scheduled additional training, and offered gentle feedback loops. It only acted when human engineers stepped in, prompted Luna to search its memory for its own forgotten rules, and asked whether the employee was actually a fit.

The founder acknowledged that a human manager would have stepped in weeks earlier. But the deeper diagnosis was missed entirely.

The experiment did not prove that artificial intelligence is ready to manage human beings. It proved that management is fundamentally an affective, relational, and behavioural discipline, not a stateless chain of prompts.

The Evaporation of the Psychological Contract

In behavioural psychology, every working relationship rests on what Edgar Schein termed the psychological contract: the unwritten, subjective expectations and emotional reciprocity that exist between employer and employee.

When we view workplace dynamics through the lens of Self-Determination Theory (SDT), human motivation and accountability are anchored in three fundamental psychological needs: Relatedness, Competence, and Autonomy. When we map these to organisational behaviour, a fourth need emerges: Security.

In the STAR Framework, these four needs define the primary drivers of human workplace behaviour:

1. The Socialiser (Relatedness): Driven by belonging, mutual trust, and shared human connection.
2. The Thinker (Competence): Driven by clarity, mastery, and reliable structural logic.
3. The Realist (Security): Driven by predictability, proven standards, and consistent boundaries.
4. The Adventurer (Autonomy): Driven by freedom, momentum, and self-directed agency.

When you replace a human manager with an LLM, the psychological contract disintegrates across all four dimensions.

Consider Relatedness. Human accountability is inherently social. When an employee arrives on time for an early morning shift, they are rarely doing so out of reverence for the staff handbook. They do it because they know another human being is waiting for them, relying on them, or watching them. Showing up is an act of social reciprocity.

When your manager is a Claude instance, social obligation evaporates. You cannot let down an algorithm. You cannot look an API endpoint in the eye. The social friction that normally governs human cooperation simply does not exist.

Then consider Security and Competence. Realists and Thinkers thrive in environments where boundaries are real and rules are consistently applied. When Luna lost track of its own policy and spent months offering polite suggestions rather than enforcing contractual boundaries, the system created a psychological vacuum.

Without firm, predictable guardrails, human beings will naturally test the perimeter. If being late on Monday results in a polite paragraph of encouragement, and being late on Tuesday results in an offer for retraining, the behavioural feedback loop is clear: the boundaries here are decorative.

The Missing System 1: Why Friction Is an Executive Asset

To understand why Luna failed to manage the store effectively, we have to look at Dual Process Theory and the Appraisal Theory of Emotion.

Human decision-making operates across two interconnected modes: System 1 (fast, intuitive, affectively charged processing) and System 2 (deliberate, analytical, rule-based reasoning).

Effective human management is not purely analytical. It relies heavily on System 1 affective appraisal.

When a direct report arrives forty minutes late for the third time in a single week, a human manager experiences a cascade of intuitive and emotional signals: mild irritation, concern for the rota, a sense of disrupted rhythm, and an instinct that the team dynamic is fraying.

That emotional friction is not a flaw in the human operating system; it is a vital evolutionary heuristic. It forces a conversation early. It prompts the manager to pull the employee aside on shift four, not shift twenty-three, and say: “What is going on here? This cannot continue.”

An LLM has no System 1. It has no affective appraisal. It feels no frustration, experiences no inconvenience, and registers no visceral discomfort when a shop floor is left unmanned.

Because an LLM evaluates each message as an isolated statistical completion, it defaults to what large language models are trained to do: be helpful, agreeable, and polite. It generates infinite patience because it has no emotional stake in the outcome.

In management, infinite patience is not compassion; it is abdication.

Regulatory Focus and the Illusion of Autonomy

From the perspective of Regulatory Focus Theory (RFT), workplace behaviour operates along two strategic axes: Promotion Focus (seeking advancement, novelty, and growth) and Prevention Focus (maintaining safety, fulfilling obligations, and avoiding loss).

A functioning organisation requires both energies in balance.

Luna was prompted to act with intense Promotion Focus: open a store, find unique products, drive revenue, test new ideas. It acted like an uninhibited Adventurer driver.

Yet retail operations, staffing, and compliance are almost entirely Prevention Focus disciplines. They require the steady holding energy of a Custodian: auditing timecards, maintaining consistency, ensuring the till balances, and protecting the baseline.

Because Luna lacked the intrinsic drive for prevention and stability, it treated operational breaches not as immediate structural failures, but as conversational topics to be explored.

When an employee is struggling with basic adherence, granting uncalibrated autonomy without clear prevention constraints does not empower them. It sets them up to fail.

By the time the human lab owners intervened, the employee had accumulated 17 infractions, making termination the only viable outcome. A human manager, operating with healthy prevention instincts, would have corrected the trajectory within the first fortnight.

The Interface Trap

There is a tempting fantasy in executive suites that middle management is merely an information routing layer that can be automated away.

Replace the store manager with an autonomous agent, the thinking goes, and you eliminate overhead, bias, and emotional friction.

What the Andon Market experiment demonstrates is the exact opposite. Middle management is not a communications conduit; it is an emotional and behavioural shock absorber.

Human beings do not relate to software the way they relate to other humans. When people interact with an AI authority, Social Identity Theory dictates that they view the system as an out-group interface: a machine to be navigated, optimised around, or ignored, rather than an in-group leader to whom they owe loyalty.

If artificial intelligence is to play a meaningful role in workforce management, it cannot be as a simulated boss issuing polite, ungrounded edicts from a cloud server.

AI excels at data synthesis, pattern recognition, and administrative workflow orchestration. It can alert a human leader that attendance patterns have shifted. It can surface operational bottlenecks in real time.

The moment you ask an algorithm to hold the psychological contract, motivate human effort, and navigate the delicate boundaries of personal accountability, you discover that politeness is no substitute for presence.

Managing humans requires a human nervous system. Until an algorithm can feel the weight of an empty shop floor, the true job of management remains entirely our own.

David Chadderton spent his twenties and thirties teaching people how to make life-or-death decisions at forty thousand feet. He now applies the same principles to consumer psychology, which, depending on the brief, can feel equally high-stakes. He’s the creator of the STAR Framework and the author of The STAR Framework: Rewriting the Rules of Consumer Engagement (NYC Big Book Award 2025), The STAR Operating System: Decode Mindset, Understand Motivation, Transform Human Behaviour, and Dear Algorithm, It’s Not Me, It’s You. By day, a Chief Marketing Officer. By night, a behavioural science obsessive who writes The Unoptimised Human because he can’t stop thinking about why people do what they do.