AI won't break the Peter Principle. It will obscure it.

The Peter Principle began as satire but rests on a deceptively simple observation: people rise to their level of incompetence. Get promoted for being good at something, land in a role that requires a different set of skills, fail to develop them, and stay stuck there. It has always raised the same uncomfortable question: how do you actually tell who is good at their job?

For a long time, output was the answer. Not a perfect answer, but a coherent one. If someone consistently produced strong work, you inferred they understood what they were doing and should be doing more of it. You promoted them.

AI is making that inference unreliable.

The signal is degrading

What is changing is not the existence of a mismatch between role and capability. It is the visibility of that mismatch.

AI has raised the baseline quality of output across knowledge work. A 2023 study by Shakked Noy and Whitney Zhang at MIT found that access to generative AI significantly improved both speed and quality of work, with the largest gains concentrated among lower-performing workers, effectively compressing visible differences in skill. (Source)

When a deliverable is strong, it is no longer clear how much of that strength comes from the person and how much comes from the system supporting them. Organizations are still reading output the way they always have, as a direct reflection of individual capability. That interpretation is increasingly wrong.

Employees appear ready for the next level earlier in their careers because the work they produce contains fewer visible gaps. Communication is tighter, deliverables feel more complete, the distance between junior and senior output narrows at the surface. What looks like accelerated development is often a combination of real improvement and tool-driven augmentation, and most organizations cannot distinguish between the two.

Where the gap shows up

The consequences tend to surface one level up, where the nature of the work changes.

At higher levels, the job shifts away from producing answers and toward defining problems, setting priorities, and making tradeoffs under uncertainty. Erik Brynjolfsson and others have argued that AI increasingly shifts value away from routine execution and toward higher-order judgment. (Source) That aligns with what many organizations are experiencing in practice.

The challenge is that judgment is not built through polished outputs. It is built through exposure to ambiguity, iteration, and failure over time. If earlier stages of a career are increasingly mediated by tools that compress or bypass those experiences, the development of judgment becomes uneven in ways that are hard to see until it matters.

There is also a well-documented cognitive risk embedded in this dynamic. Research on automation bias shows that people tend to over-rely on automated systems, often accepting outputs without sufficient critical evaluation. (Source) In a workplace context, this does not just affect individual decisions. It shapes how capability develops over time. People become more efficient while also becoming less practiced in independent reasoning, and that tradeoff is easy to miss when output continues to look strong.

The mismatch goes latent

What makes this particularly difficult to manage is that AI can mask these gaps for longer than previous systems could. A manager can continue producing high-quality work with tool assistance. An executive can rely on structured synthesis to support decision-making. From the outside, performance looks stable, even strong. Internally, capability may be more fragile than it appears, because it depends on scaffolding that is not always visible and not always transferable.

When tools raise the baseline quality of output across the board, with the largest gains among lower performers, the distance between someone who genuinely has the judgment for the next level and someone who doesn't starts to disappear at the surface: deliverables look complete, communication feels tighter, readiness looks accelerated. The principle evolves from a visible ceiling into a delayed realization.

But judgment isn't built through polished outputs. It's built through ambiguity, iteration, and failure over time. If those experiences are being compressed or bypassed, the gap doesn't disappear. It goes latent.

And latent gaps are more expensive than visible ones. The mismatch surfaces later, further up the hierarchy, when the stakes are higher and the cost of correction is greater.

What organizations actually need to change

The implication is not that AI should be resisted, but rather that the criteria used to evaluate people need to change. As Peter Drucker famously noted, what organizations choose to measure shapes what they reward. If measurement systems remain anchored in output alone, they will continue to reinforce the distortions AI is introducing.

At the end of the day, this is not a new problem, accelerated. Output was never the whole story, and it's even less of the story now. A few things worth considering and designing around within any organization:

• ‍Separate the signal from the scaffold. When evaluating someone's readiness for the next level, look at what they produce without AI support, not just what they produce with it. Ask them to walk you through their reasoning rather than their output. The thinking is the thing.‍
• Assess for judgment in unstructured situations. The clearest test of whether someone is genuinely ready for a more senior role is how they operate when there is no clear prompt, no structured framework, and no obvious answer. Build those moments into your evaluation process deliberately.‍
• Reconsider what you are promoting for. Execution is becoming more abundant, so discernment is the main constraint. The ability to decide what matters, prioritize effectively, and navigate ambiguity is not being automated at the same pace. If your promotion criteria were built around output volume and delivery speed, they need to be rebuilt around something closer to the source of actual value needed and delivered.‍
• Watch for scaffolding dependency at the leadership level. The most expensive version of this problem is a senior leader whose performance depends entirely on tools and systems that cannot be transferred. Run the diagnostic before that person is in a role where the cost of discovering it is high.