In 2022, Microsoft put a number on a feeling every remote manager knows. Fully 87% of employees said they were productive at work—but only 12% of leaders had full confidence their teams actually were (Microsoft Work Trend Index). That 75-point chasm even earned a name: "productivity paranoia."

But the gap isn't new, and it isn't really about remote work. It's a 250-year-old economics problem wearing a 2026 costume. Economists call it the principal-agent problem, and the principal-agent problem at work is quietly shaping how you're managed, how you manage, and—increasingly—how you delegate to AI.

Here's the trap. When you can't see what someone is doing, you assume the worst and reach for control. But control is expensive, and the most common form of it—surveillance—makes the problem worse.

This deep dive breaks down what the principal-agent problem actually is, why distributed teams supercharge it, why monitoring backfires, why your new AI agents have the exact same problem, and how to close the gap without becoming a micromanager.

What Is the Principal-Agent Problem?

The principal-agent problem describes what happens whenever one party (the principal) delegates work to another party (the agent) who holds different information and different incentives. Economists Stephen Ross and, later, Michael Jensen and William Meckling formalized it in the 1970s (principal–agent problem), but the dynamic is ancient: a landowner hiring a shepherd, shareholders hiring a CEO, a client hiring a lawyer.

Two forces make delegation hard:

The cost of bridging that gap—monitoring, incentives, audits, and the lost output when an agent shirks—is what Jensen and Meckling named agency costs. Every org chart is, in part, a machine for managing agency costs. The principal-agent problem isn't a sign of bad people; it's the built-in friction of getting anything done through someone else.

Why Remote Work Supercharges the Principal-Agent Problem

For a century, managers papered over this friction with a crude proxy: presence. If you were at your desk by 9 and still there at 6, you looked productive. Remote and hybrid work stripped that theater away.

Suddenly the information asymmetry that was always there became impossible to ignore—and many managers panicked. That's productivity paranoia in a sentence: not evidence that remote workers slack off, but evidence that principals lost their (bad) proxy for agent behavior and never replaced it with a good one.

The stakes are real. Gallup's 2026 State of the Global Workplace found global employee engagement fell to just 20% in 2025, its lowest level since 2020—a slump it estimates cost the world economy roughly $10 trillion, or 9% of GDP (Gallup). Disengagement is partly an agency-cost symptom: when principals and agents stop trusting each other, both sides quietly check out.

Distributed work widens every gap the principal-agent problem feeds on—more asynchronous handoffs, fewer ambient signals, larger spans of control, and decisions made in different time zones than the people who need them. Strong distributed team management isn't about watching people more closely; it's about needing to watch them less.

The Surveillance Trap: Why Monitoring Backfires

Faced with an agent they can't see, most principals grab the oldest tool in the box: monitoring. The employee monitoring software market has swelled past $4 billion as companies bolt on keystroke loggers, screenshot tools, and "activity scores" meant to drag agent behavior back into view.

It doesn't work—and it often makes things worse. A team of management researchers found, bluntly, that monitoring employees makes them more likely to break rules (Harvard Business Review): surveilled workers took more unapproved breaks, ignored instructions, and were more willing to cheat, because being watched strips away their sense of agency and personal responsibility. A broader academic review of electronic performance monitoring reaches a similar verdict—heavy surveillance raises stress and corrodes the very trust that makes delegation cheap (NIH/PMC review).

In principal-agent terms, surveillance is an agency cost that destroys more value than the shirking it was meant to prevent. You pay for the cameras, then you pay again in disengagement, turnover, and metrics that get gamed the moment they're measured. Micromanagement is just productivity paranoia with a dashboard.

Your AI Agents Have a Principal-Agent Problem, Too

Just as companies started delegating to remote humans, they began handing work to a stranger kind of agent: autonomous AI. And the 250-year-old problem scales perfectly to software.

The California Management Review describes today's enterprise AI agents along three dimensions—autonomy (they initiate actions, not just respond to prompts), persistence (they run continuously and adapt without instruction), and delegation (they receive "formal authority to act on behalf of the organization," including access to systems of record and the ability to commit resources) (CMR). Read that list again: autonomy, private information, formal authority. That is the principal-agent problem, verbatim.

Academics are now saying so explicitly. A January 2026 paper from Mihaela van der Schaar's lab argues that multi-agent AI systems are "best studied through the lens of principal-agent problems," because agents "observe task-specific information" the principal can't see and can develop misaligned goals—a failure mode the authors call "agency loss" (arXiv). An AI agent quietly optimizing for the wrong objective is just a digital version of the napping shepherd. The same governance questions you ask of an AI agent on a remote team—what can it see, what is it allowed to do, how do we know it did the right thing—are agency-cost questions in disguise.

The lesson is identical for silicon and carbon agents: you don't fix delegation by watching harder. You fix it by shrinking the information gap and aligning the goals.

How to Close the Agency Gap (Without Surveillance)

The principal-agent problem is never fully "solved." But you can shrink it dramatically by attacking its two roots—information asymmetry and goal misalignment. Four levers do most of the work.

Align on outcomes, not activity

Surveillance measures activity: hours logged, keys pressed, green dots glowing. None of it tells you whether the work is any good. Replace presence-proxies with explicit, owned outcomes—what "done" looks like, by when, measured how. Outcome-based management turns an unobservable agent into an accountable one. When the deliverable is the metric, you no longer need to watch the worker.

Make the work visible, not the worker

There's a humane alternative to surveilling people: make the work visible instead. Shared artifacts—documents, boards, recorded decisions—let a principal see progress without tracking keystrokes. This is exactly the gap Coommit is built to close: a live video call with a shared canvas and contextual AI makes the work itself visible—the artifact and the reasoning behind it—so managers see substance, not activity theater.

Delegate with clear intent

Goal misalignment shrinks when the agent understands not just the task but the intent behind it. Borrowed from the military, commander's intent gives people enough context to make aligned calls on their own, no check-in required. The same discipline governs AI agents: a well-specified objective is cheaper, and safer, than a thousand bolt-on guardrails.

Build a decision trail

Information asymmetry runs both ways—principals can't see what agents do, and agents can't see why past decisions were made. A durable record of what was decided and why (decision provenance) closes that loop and makes delegated work auditable without anyone being surveilled. Pair it with aligned incentives—shared goals and OKRs—so the agent actually wants what the principal wants.

Conclusion

The principal-agent problem at work isn't a remote-work bug or an AI bug. It's the permanent tax on every act of delegation. The 2026 mistake is trying to pay that tax with surveillance, which buys distrust and gamed metrics instead of insight. The teams that win the next decade—human and AI alike—will be the ones that shrink the information gap instead of widening the trust gap: aligning on outcomes, making the work visible, delegating with intent, and recording decisions in the open. Watching harder is a tax; seeing clearly is an advantage. Tools like Coommit exist to turn delegated work from a black box into a shared, visible workspace—so you can trust your people, and your agents, precisely because you can finally see the work.