The Productivity Paranoia Trap

TL;DR

• The trust gap is now structural. 85% of leaders say hybrid work makes it hard to trust employees are productive. 87% of employees say they are productive. The 2-point gap between those numbers is not a rounding error — it's a trust collapse that has hardened into workplace architecture.
• AI is being deployed on both sides of the gap. Employers are installing keystroke tracking, screen capture, and "productivity scoring" algorithms. Employees are using unsanctioned AI tools to meet rising expectations — and hiding it. 75% of employees say they'd trust AI over their manager for at least one task.
• The monitoring is not about productivity. Meta's CTO told employees there is "no way to opt out" of keystroke tracking — but the data won't be used in performance reviews. It will be used to train AI agents to replace the workers being tracked. The surveillance is training data.
• The mental health toll is measurable and accelerating. 72% of employees say their employer prioritises productivity over wellbeing (up 11 points in one year). 52% have used substances to cope with work stress during the workday. 51% have cried due to work stress in the past 30 days.
• The cycle is self-reinforcing. Monitoring drives presenteeism. Presenteeism drives burnout. Burnout drives turnover. Turnover drives more monitoring. Breaking it requires a manager willing to say: "I don't need to watch you work. I need to see what you produce."

What Happened

Three data points landed in the past month that, taken together, describe a workplace transformation that is not being named.

First: Meta announced it would install tracking software on every US employee's computer. The program, called the Model Capability Initiative, captures keystrokes, clicks, mouse movements, and screenshots across work applications including Gmail, GChat, and the company's internal AI assistant. When employees asked how to opt out, CTO Andrew Bosworth replied: "There is no way to opt out on your work laptop." Staff reacted with shocked, crying, and angry emoji ¹.

The purpose is not performance evaluation. Meta explicitly stated the data would not be used in reviews and managers would not see it. The purpose is to train AI agents — to capture "real examples" of how humans interact with computers so that Meta can build systems that replace those interactions. As Bill Howe, associate professor at the University of Washington, put it: "Employees everywhere are helping to train the systems that will take their jobs" ¹.

Meta is spending $135 billion on AI infrastructure this year. It is also laying off 8,000 employees — 10% of its workforce — starting May 20. The tracking software and the layoffs are not separate stories. They are the same story.

Second: a survey of 1,200 C-suite executives and 1,200 employees by Writer and Workplace Intelligence revealed that the AI adoption story most companies are telling is fiction. Seventy-five percent of C-suite respondents said their company's AI strategy is "more for show" than for actual internal guidance. Thirty-nine percent admitted they have no formal plan to drive revenue from AI tools. Nearly 70% said that while that foundation is missing, their company is already doing layoffs tied to AI ².

The survey also found that 60% of executives are planning layoffs for employees who "can't or won't use AI." Forty-four percent of Gen Z workers admit to sabotaging AI in at least one way — intentionally generating low-quality outputs to make AI look ineffective, tampering with performance metrics. Thirty percent say they don't want AI to take over their job. Twenty-six percent say it has diminished their value or creativity ².

Employee confidence in their company's AI strategy fell from 47% in 2025 to 31% in 2026. And 75% of employees say they'd trust AI over their manager for at least one task ².

Third: Modern Health's annual workplace mental health survey documented a collapse in employer trust that is accelerating. Just 33% of employees strongly agree that their employer values their mental health, down from 41% in 2025. Seventy-two percent say their employer actively encourages productivity at the expense of personal wellbeing, up from 61% — an 11-point increase in a single year ³.

The AI-specific findings are stark: 69% of employees believe AI will lead to layoffs at their own company within three years. Forty-nine percent are personally afraid of losing their job to AI. Twenty-four percent say AI is already negatively affecting their mental health. Sixty-seven percent say AI has raised productivity expectations, and among that group, 64% report increased stress as a direct result ³.

The coping mechanisms are alarming: 63% report using alcohol, THC, or unprescribed pharmaceutical drugs to relieve stress after work. Fifty-two percent have used substances during the workday itself. Among Gen Z, THC has overtaken alcohol as the primary after-work coping mechanism, and daytime use (51%) nearly matches after-work use (59%) ³.

These three data points are not separate stories. They are the same story, viewed from three angles: the employer, the employee, and the body.

What It Actually Means

The conventional narrative about AI and the workplace goes like this: AI makes workers more productive. More productive workers make companies more profitable. Everyone wins, once we get through the transition.

The data says something different. It says that AI is being deployed simultaneously as a productivity tool and as a surveillance tool — and that the surveillance function is not a side effect. It is the point.

Consider the Meta case carefully. The company is not tracking keystrokes to measure productivity. It is tracking keystrokes to capture training data for AI agents that will eventually perform the work being tracked. The surveillance is not about monitoring the worker. It is about replacing the worker. The worker is generating the training data for their own replacement, on a device they cannot opt out of, at a company that is simultaneously laying off 10% of its workforce.

This is not a conspiracy theory. It is disclosed in an internal memo, confirmed by the CTO, and reported by Reuters and CNET. The memo says: "This is where all Meta employees can help our models get better simply by doing their daily work" ¹.

Now layer on the Writer survey finding: 75% of C-suite executives say their AI strategy is "for show." Nearly 70% are doing AI-driven layoffs without a plan to generate revenue from AI. Sixty percent are planning to fire people who won't use AI. The strategy is not a strategy. It is a posture. And the posture is: adopt AI or leave.

The employee response is rational. If your employer is tracking your keystrokes to train your replacement, and simultaneously telling you that you must use AI or be fired, and simultaneously admitting that the AI strategy is "for show" — what is the rational response? For 44% of Gen Z, the answer is sabotage. For 75% of all employees, the answer is trusting AI more than their manager. For 52%, the answer is substances during the workday.

The trust gap — 85% of leaders don't trust employees to be productive, 87% of employees say they are productive — is not a measurement problem. It is a structural feature of a workplace where the tools of production have become tools of surveillance, and the tools of surveillance have become tools of replacement.

Hype Deconstruction

Three things this story is not:

It is not about remote work. The productivity paranoia narrative is often framed as a hybrid-work problem: managers can't see employees, so they don't trust them. But Meta's tracking software is installed on devices in the office and at home. The surveillance is ambient. It doesn't matter where you are. What matters is that your interactions are being captured as training data. The location of the worker is irrelevant to the architecture of the system.

It is not about a few bad companies. Meta is the most visible example, but the pattern is industry-wide. Nvidia expects engineers earning $500,000 or more to spend at least $250,000 on AI tokens — and considers anything less "cause for alarm" ⁴. Meta requires a percentage of code changes to be "agent-assisted," factored into performance reviews ⁴. McKinsey operates 25,000 AI agents alongside 40,000 employees and expects parity soon ⁴. One Silicon Valley founder gives employees full AI access, then cuts it off — and puts anyone who doesn't beg for it back on a performance improvement plan ⁴. The "tokenmaxxing" trend — measuring employee value by AI token consumption — is spreading beyond Silicon Valley ².

It is not a temporary transition. The mental health data suggests the strain is compounding, not resolving. Trust in employers fell 8 points in one year. The belief that employers prioritise output over people rose 11 points. The percentage of employees who have cried due to work stress rose 12 points. These are not transition pains. They are trend lines.

Stakeholder Landscape

Who is directly affected: Knowledge workers at large technology companies are on the leading edge — Meta's 79,000 employees, the 40,000 at McKinsey, the engineers at Nvidia. But the Writer survey data covers 1,200 employees across the US, UK, Ireland, Benelux, France, and Germany. This is not a Silicon Valley story. It is a white-collar story.

Who is second-order affected: Managers. Eighty-two percent say being a manager is harder than ever. Only 37% feel equipped to identify burnout in their teams. Forty percent of senior managers received a new mental health diagnosis in the past 12 months — more than three times the rate of non-managers. Senior managers report the highest levels of AI anxiety: 74% expect AI-driven layoffs, and 57% personally fear for their own job ³. The people implementing the surveillance are also afraid of being replaced.

Who benefits from the noise: AI infrastructure companies. The $135 billion Meta is spending on AI capex this year flows to Nvidia, to data centre operators, to cloud providers. The surveillance software vendors — Teramind, the "Model Capability Initiative," the "productivity scoring" startups — are building a new category on the trust gap. The larger the gap, the larger the market.

Who is not affected despite the noise: Frontline workers whose jobs cannot be captured as keystroke data. The surveillance architecture is designed for knowledge work. If your job involves physical presence — nursing, construction, food service, logistics — you are not being tracked this way. You may be tracked in other ways, but the AI-training-surveillance pipeline described here is specific to screen-based work.

Cross-Layer Implications

The talent market is bifurcating. The Writer survey identifies a "two-tiered workforce": AI super-users save nine hours per week and are three times more likely to have received a promotion and a raise. Everyone else is being told to catch up or be fired. The gap between super-users and everyone else is not closing — it is widening, and it is being formalised in performance review systems.

The regulatory layer is missing. There is no federal framework in the United States governing the use of employee surveillance data for AI training. Meta's CTO can say "no way to opt out" because there is no law that says otherwise. Eric Null of the Center for Democracy & Technology called Meta's program one of the most "invasive" forms of workplace surveillance and noted it "can cause real harm to people with disabilities" ¹. The legal architecture to address this does not exist.

The mental health infrastructure is inadequate — and employees know it. Fifty-eight percent of employees say they feel safer talking to a chatbot about their mental health than their workplace HR department, up from 50% in 2025. Fifty percent don't use mental health days at all — not because they don't need them, but out of fear of judgment. Eighty-nine percent say more mental health benefits are needed ³. The employer-provided mental health system is losing legitimacy at the same moment the need for it is rising.

The investor incentive structure rewards the behaviour. Meta's stock did not fall when the keystroke tracking was announced. The layoffs — 8,000 people, 10% of the workforce — are happening alongside $135 billion in AI infrastructure spending. The market is pricing the AI transition as value-creating, not value-destroying. As long as that holds, the surveillance-replacement cycle will accelerate.

What This Means for You

If you are a knowledge worker: The question is not whether your employer will adopt AI. The question is whether your employer is using AI to augment your work or to capture your work as training data for your replacement. The distinction matters. If your company is installing tracking software, ask what the data is used for. If the answer is "training AI agents," understand that you are generating the asset that will compete with you. The rational response is not sabotage — it is to understand which of your skills are not capturable as keystroke data and to invest in those.

If you are a manager: The trust gap is your problem, not HR's. Your team knows whether you are measuring output or activity. If you are measuring activity — keystrokes, screen time, mouse movements — you are training your team to perform activity rather than produce output. The alternative is harder: define what good work looks like, measure that, and accept that you cannot see the process. But the alternative is the only thing that breaks the cycle.

If you are an executive: The Writer survey finding — 75% of C-suite say their AI strategy is "for show" — should be alarming. Your employees know. They are not confused about whether the strategy is real. They are responding to the gap between what you say and what you do. If you are doing AI-driven layoffs without a revenue plan, you are not executing a strategy. You are managing headcount and calling it transformation. The market may reward that in the short term. Your workforce will not.

If you are a policy-maker: There is no legal framework in the United States governing the use of employee surveillance data for AI training. The EU's AI Act and GDPR provide some guardrails, but the US has nothing equivalent. The Meta case — "no way to opt out" — is a preview of what happens when the regulatory layer is absent. This is not a technology problem. It is a governance problem.

Uncertainty Ledger

What's still unresolved:

• Will the Meta model spread? Meta is the most visible example, but the Writer survey suggests the AI-surveillance-posture is already widespread. The question is whether non-tech companies adopt the same architecture — keystroke tracking as AI training data — or whether that remains specific to companies building AI agents.
• Will the mental health trend lines bend? The Modern Health data shows acceleration, not stabilisation. The question is whether there is a floor — a point at which trust cannot fall further — or whether the trend continues until some external shock (regulation, unionisation, a high-profile lawsuit) intervenes.
• Will the "two-tiered workforce" become permanent? The Writer survey identifies AI super-users as a distinct class. The question is whether the gap between super-users and everyone else narrows as AI tools become more accessible, or widens as the super-users accumulate compounding advantages.
• What would change the analysis: A federal employee surveillance law. A successful class-action lawsuit against non-consensual workplace tracking. A reversal in the market's reward for AI-driven headcount reduction. Any of these would alter the incentive structure that is currently driving the cycle.

Bottom Line

The productivity paranoia gap — 85% of leaders don't trust employees, 87% of employees say they're productive — is not a measurement problem. It is a trust problem that has been architectural: AI surveillance on one side, AI shadow-use on the other, and a widening chasm where management used to be. The companies deploying keystroke tracking are not trying to close the gap. They are using the gap as training data. The cycle — monitor, replace, monitor more — will not break on its own. It will break when someone with authority decides that measuring output is harder but better than measuring keystrokes, and that trust is not a sentiment. It is a system design choice.

Sources