Featured
Cheap Execution, Expensive Judgment
What happens when nobody learns the work?
“Earn while you sleep” is one of those lines you hear in founder circles, shorthand for leverage. Build something that keeps producing even when you’re not there.
Naval Ravikant popularized the phrase, but the logic now extends beyond assets or products. It applies to execution itself. We can produce certain kinds of work continuously at low marginal cost: drafts, summaries, analyses, proposals, customer replies, and code suggestions. The system routes tasks through synthetic capacity and returns outputs that are often usable, sometimes excellent, and occasionally wrong in ways that are difficult to spot. It does this without fatigue, without junior analysts grinding through first passes, and without the slow apprenticeship that used to sit behind professional fluency.
That scaling power is real. It is also easy to misunderstand.
If you’ve read my earlier piece, The Employee Who Never Sleeps, you’ll recognize the premise. This essay is about the next consequence: what happens to judgment when execution becomes cheap.
The usual conversation circles around replacement. Will jobs disappear? Will whole functions shrink? Those questions matter, but they are not the only ones. The quieter effect is this: when execution becomes cheap, judgment becomes the scarce resource, and the mechanisms that create judgment quietly weaken.
For most organizations, the formation of competence was not a mystery. It was procedural. Juniors did junior work. They handled the first drafts, the reconciliations, the background searches, the client follow-ups, the tedious checks. They learned the standards and the edge cases. They made small errors and were corrected. They built a sense of what “good” looks like in this firm, with these clients, in this industry, under these constraints. Over time, they earned the right to make bigger calls.
That ladder was not always admirable. It was often inefficient, sometimes exploitative, and occasionally padded with nonsense. Still, it served a purpose. It created judgment through repetition, proximity to consequences, and exposure to the boring details that professionals later claim to “just know.”
Synthetic labor compresses that ladder.
If a system can produce a plausible first draft in minutes, the temptation is to remove the human first draft altogether. If a system can generate ten options instantly, the temptation is to turn deliberation into selection. If the system can prepare the client deck, the temptation is to treat presentation as the work.
This is where we meet what I think of as the practice deficit. The organization looks more productive. It may even be more productive in the narrow sense of shipping outputs. But the repeated doing that once formed competence starts to disappear from human hands. The implication is not that nobody will be skilled. It is that skill will become thinner, more uneven, and more fragile under novelty.
Consider legal research. A junior associate at a law firm once spent weeks reading case law, building a sense of precedent, and learning to distinguish strong citations from weak ones. That repetition built judgment. Now, AI retrieves relevant cases in minutes. The senior reviews the output, but the junior never touched the material. The deck gets filed. The case moves forward. But the associate who skipped that apprenticeship later struggles to spot when an AI-generated memo mischaracterizes a holding or omits a key distinction. The fluency looks the same. The foundation does not.
Michael Polanyi’s point about tacit knowledge still holds: we know more than we can easily say. Much of professional judgment is not a checklist. It is the ability to notice what is missing, to sense when a confident answer is too smooth, to recognize a pattern that does not quite fit. That kind of knowing is built. It is not downloaded.
When execution is cheap, verification becomes more important. It also becomes harder to do well.
Herbert Simon warned that bounded rationality makes attention the scarce resource inside organizations. That observation becomes painfully practical when synthetic capacity floods the workplace with plausible material. Production accelerates. Attention does not. Review becomes the bottleneck, but review is precisely the activity many firms under-invest in because it feels like overhead and slows the tempo.
Get Nicos Rossides’s stories in your inbox
Join Medium for free to get updates from this writer.
We then get a quiet inversion. In the old model, juniors produced work and seniors reviewed it. Review was an act of teaching as well as control. In the new model, humans may be asked to judge outputs they did not produce, at a speed set by the tool, without the depth of practice that once made judgment reliable. We can see how that creates risk. Not only of errors, but of a certain kind of hollow expertise.
By hollow expertise, I do not mean incompetence. I mean fluency without load-bearing understanding. A team can sound confident, move quickly, and deliver polished outputs while lacking the internal grip that shows itself when circumstances change. Under routine conditions, this may not matter. Under ambiguity, pressure, or tool failure, it matters a great deal.
It also changes how organizations learn. James March’s distinction between exploitation and exploration is useful here. Synthetic capacity can turbocharge exploitation, the rapid reuse of known patterns. It can also starve exploration, the slow work of learning when the pattern breaks. If the firm optimizes for speed and volume, it may gradually become less able to deal with problems that do not resemble yesterday’s.
This is not an argument that disaster is inevitable. Some firms will use synthetic capacity to raise standards, not lower them. Some will build stronger verification cultures. Some will protect apprenticeship intentionally. The point is simpler: cheap execution changes the economics of work, and the formation of judgment does not automatically keep up.
That is why governance becomes the central question, not as a slogan but as an operational design problem.
These are not bureaucratic luxuries. They are the operational minimum for an organization that wants to scale execution without eroding competence. Without them, cheap production becomes expensive failure, just on a delayed timeline.
The first requirement is auditability in the literal sense. An operational record of work should show what was routed through synthetic systems, what inputs were used, what outputs were accepted, what edits were made, and by whom. If the work cannot be reconstructed, it cannot be owned. If it cannot be owned, it cannot be learned from. Many organizations will discover that they have adopted speed while keeping traceability at the level of polite fiction.
Second is clear ownership at the decision point. “The team” is not ownership. “The tool” is not ownership. If an output becomes consequential, a named role owns it at the moment it becomes consequential, regardless of how it was produced. This is not moral drama. It is how feedback loops function. Without ownership, errors become atmospheric. Everyone senses them, nobody holds them.
Third is escalation rights that are real in practice. When a person reviewing synthetic output is uncertain, there must be a route upward that is legitimate and safe. In many workplaces, escalation is treated as weakness, or as slowing things down. Cheap execution makes that cultural habit expensive. If production is abundant and review is scarce, punishing escalation is a way of guaranteeing that weak signals are ignored until they become failures.
Fourth is protected human veto authority. Not a ceremonial “human in the loop,” but a genuine right to pause a workflow, stop a release, or refuse an output without requiring political heroics. A veto that cannot be used under pressure is not a veto.
None of this requires regulatory theatre. It requires clarity. It requires the firm to admit that it is trading one constraint for another. Execution becomes abundant. Judgment becomes scarce. If you want the gains, you have to pay for the scarce part.
Which brings us back to leverage.
Ravikant’s phrase “permissionless leverage” is useful precisely because it is incomplete. Leverage may be easier to access. Responsibility is not. In the workplace, the seductive move is to treat synthetic capacity as a productivity upgrade with no downstream obligations. The obligation is not abstract. It is the redesign of training and accountability.
If you want one crisp takeaway: permissionless leverage is not permissionless responsibility.
That line matters because the risk is not only that work gets automated or jobs shift. The deeper risk is that organizations start to look like they’re gliding while the muscles that once kept them stable quietly weaken. If execution scales, judgment has to be trained and accountability has to be assigned in ways that survive speed. The closing question is not whether the outputs feel real or the process looks effortless. It’s what it costs to make work look effortless, and who pays that cost. If the payment is made in missing judgment and diffused responsibility, the leverage was never free. It was merely deferred.
Dr Nicos Rossides is a CEO, educator, and author exploring the intersection of management, philosophy, and technology. His work bridges practice and theory, exploring how intelligent technologies are redefining leadership, governance, and the future of work. His recent books include The Future of Work: Managing in the Age of AI (2024) and Employee Engagement in Startups: Navigating Growth, Culture, and Innovation (Routledge, 2025).
