Two companies buy the same workplace AI platform this quarter. The first configures it to flag idle time, score keystroke activity, and alert managers when response latency rises. The second configures it to remove status reporting, surface blockers, and track whether the work shipped. Identical software, opposite philosophies, and within a year they will be different companies. Every AI deployment in the workplace is, underneath the procurement language, an answer to one question: do we fundamentally believe our people want to work, or do we believe they must be made to?
The question is old enough to have a founding story. At 25, as Ricardo Semler has told it, he collapsed on a factory floor and was told by a doctor that he was the healthiest sick person the doctor had ever seen: nothing wrong except the way he lived and worked. That afternoon he began dismantling the apparatus, the org chart, the time clock, the six-day week, and eventually the deeper habit underneath them: counting hours instead of results. Semco stopped tracking time and started asking whether the pump got delivered. Absenteeism fell. The lesson was not that measurement is bad; Semco measured relentlessly. It was that measuring activity is a confession that you cannot, or will not, measure outcomes.
What surveillance actually buys
Modern monitoring tools inherit the time clock's logic with vastly better sensors, and the telemetry they generate is genuinely revealing, just not in the intended direction. Microsoft's 2025 Work Trend Index, drawing on Microsoft 365 telemetry, reports knowledge workers interrupted every two minutes during business hours, roughly 275 interruptions a day, with after-hours chats up 15% year over year. Read as a management dashboard, those numbers invite more monitoring of a workforce that seems scattered. Read honestly, they indict the system, not the workers: the always-visible workplace is already producing fragmentation and spillover, and adding surveillance intensifies the very visibility that causes it.
The deeper cost is what monitoring communicates. Mayer, Davis and Schoorman's classic model defines trust as the willingness to be vulnerable to another party, built on perceived ability, benevolence, and integrity. Surveillance is the formal refusal of vulnerability. It tells every employee, in architecture rather than words, that the default assumption is shirking. People are reliable readers of architecture. A company can print "we trust our people" on every wall; the software that screenshots their desktops is the sentence they believe.
What trust measurably buys
The trust side of the ledger is no longer soft. Johannsen and Zak, in a US sample of 1,095 working adults, found employees in the highest-trust quartile of organizations earning 10.3% more than the middle quartile, with trust correlating with productivity at r=0.55 and higher trust associated with lower chronic stress. Correlational, yes, but consistent with the cross-country evidence: Bloom, Sadun and Van Reenen found that firms in high-trust regions delegate more decision authority, and that this delegation is precisely what lets firms grow beyond the capacity of their CEO's attention. Trust is not the absence of control. It is the substitution of expensive, low-resolution control (watching) with cheap, high-resolution control (transparent outcomes owned by the people producing them).
This is what the trust-philosophy AI deployment looks like in practice: the system automates the reporting nobody should write, makes outcome data visible to the team itself before any boss sees it, and treats anomalies as questions rather than verdicts. The surveillance deployment inverts each choice: it generates reporting about people, routes it upward first, and treats anomalies as accusations pre-filed.
The choice, stated plainly
Vendors will insist the tools are neutral, and they are right in the narrow sense: the same model can summarize either a person's outputs or their minutes. That is exactly why the deployment choice is so diagnostic. Nothing in the technology forces either philosophy, so the configuration a leadership team chooses is a pure expression of its beliefs, unlaundered by necessity.
There is also a timing argument for choosing well. The Edelman Trust Barometer 2026 finds employers the most trusted institution in most people's lives, better positioned than governments or media to broker trust across a fracturing society. That standing is an asset with a balance, and algorithmic surveillance draws it down at scale, silently, one desktop screenshot at a time.
Semler's doctor had it right about companies too: the sickness is the way of working, not the workers. An organization that instruments its people to the keystroke has diagnosed the wrong patient. The one that instruments its outcomes, opens the results to everyone, and lets adults manage the hours, that one is betting the way Semco bet, and the wager has forty years of evidence and a falling absenteeism rate behind it.
Sources
- Mayer, Davis & Schoorman (1995). An Integrative Model of Organizational Trust. Academy of Management Review. Trust defined as willingness to be vulnerable; antecedents: ability, benevolence, integrity. https://www.jstor.org/stable/258792
- Johannsen & Zak (2021). The Neuroscience of Organizational Trust and Business Performance. Frontiers in Psychology. n=1,095: highest-trust quartile earned 10.3% more; trust-productivity correlation r=0.55 (correlational). https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2020.579459/full
- Bloom, Sadun & Van Reenen (2009/2012). The Organization of Firms Across Countries. NBER/QJE. High-trust regions predict greater delegation of decision authority; trust-enabled decentralization supports larger, more productive firms. https://www.nber.org/digest/dec09/international-differences-decentralization-and-productivity
- Microsoft WorkLab (2025). Work Trend Index 2025. Telemetry: interruptions every 2 minutes (~275/day); after-hours chats up 15% year over year. https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born
- Edelman Trust Institute (2026). Edelman Trust Barometer 2026. Employers are the institution best positioned to broker trust (33,938 respondents, 28 countries). https://www.edelman.com/trust/2026/trust-barometer