Jacob Burgess
Jun 10, 2026
Is AI actually making us faster?
In my last piece, I wrote about the first hidden cost of enterprise AI: overconfidence created upstream, and the downstream burden placed on experts forced to correct it. But there’s a second inefficiency emerging—quieter, more subtle, and in some ways more wasteful.
This one doesn’t come from business stakeholders misunderstanding AI. It comes from the experts themselves.
The New Pattern
A new pattern is starting to appear. An expert writes an email—clear, competent, grounded in experience. Under normal conditions, they would reread it once, make a small edit, and send. Now there’s an additional step: double-check it with AI.
The AI returns a cleaner version. Slightly more structured. Slightly more polished. Just different enough to introduce doubt. The expert compares the two versions, adjusts phrasing, reconsiders tone, and double-checks whether the original framing was actually better. The result is a loop—writing followed by second-guessing, reconciling, rewriting, and rereading. The email improves marginally, but it takes twice as long.
What makes this dynamic difficult to spot is that AI doesn’t obviously degrade quality. If anything, the final output is often slightly more polished. But the improvement is incremental, and the cost is hidden. A senior professional who once needed twenty minutes to draft and send a message now spends forty—not because the task is harder, but because the tool has introduced an optional layer of refinement that feels difficult to ignore.
Extra effort, for little gain
But that improvement is not always real. AI optimizes for structure, tone neutrality, and linguistic clarity. In legal and compliance contexts, those are not the only variables that matter. The original wording may reflect judgment, prioritization, or intentional ambiguity that a model cannot reliably preserve. The expert then has to decide whether the AI version is actually better or simply more generic. That decision takes time.
The deeper issue is that AI reintroduces uncertainty into places where expertise had already resolved it. Before AI, the expert’s internal standard determined when something was “good enough to send.” After AI, there is always an external reference point that can be consulted—and therefore ignored at some risk. Even when the answer is “no, my draft is fine,” the question itself takes time. Over the course of a day, it adds up.
This effect is most pronounced for experienced professionals. Junior employees may benefit from an AI review layer because it provides scaffolding they don’t yet have (but the inaccuracy risk increases for junior employees who don’t have the expertise to call out the AI when it is wrong). Senior experts already have that scaffolding. For them, AI is not a replacement for thinking—it is a challenge to it. Every suggestion must be evaluated, accepted or rejected, and often modified. The expert ends up doing two passes of work instead of one: generating the original output and adjudicating the alternative.
The Unintended Inefficiencies
Organizations tend to frame AI as a speed tool. For expert users, it increasingly functions as a tool for assurance. That can be valuable in high-risk situations—external communications, regulatory positioning, executive escalation. But when that same level of scrutiny is applied indiscriminately, productivity quietly declines.
The conclusion is that AI does not simply reduce work. It redistributes it:
It moves effort upstream (where overconfidence is created)
It moves effort downstream (where experts must correct it)
And now, it adds effort internally (where experts re-evaluate their own output)
None of these costs were captured in early ROI models. The solution is not to abandon AI. It is to apply it selectively. Enterprises need to evaluate who should be using AI, and when. Experts need to learn not just how to use AI, but when not to.