The AI Productivity Trap: Why "More With Less" Is the Wrong Equation

The dominant AI narrative in boardrooms right now goes something like this: deploy tools, reduce headcount, maintain output, improve margins. It's clean. It's modelable. It's probably wrong.

A new study from Orgvue cuts against the grain of that logic in a way executives should find uncomfortable. HR Dive (2025-07-14) reports that companies which hired more employees were more likely to see sustained revenue growth — even amid active AI implementation and explicit "do more with less" mandates. The implication isn't that AI doesn't matter. It's that the organizations treating this moment as a subtraction problem are systematically miscalculating it.

That miscalculation has a cost that's already showing up in operations. Employees are currently spending nearly a full workday per week — roughly 7 to 8 hours — supervising and correcting AI outputs, a phenomenon researchers are calling "botsitting," according to HR Dive (2025-07-11). That is not the productivity dividend the investment case promised. What it actually represents is a hidden labor tax: you eliminated three analysts, then quietly redistributed their oversight burden across the six you kept, who now spend Thursdays babysitting models instead of doing the work AI was supposed to free them for.

The honest diagnosis here isn't that AI tools are failing. It's that organizations are deploying them into structures that were never designed to absorb them. McKinsey's editorial position on this is blunter than you'd expect from a firm with skin in the AI consulting game: McKinsey (2025-07-10) argues that "companies that bet on equipping good staff with extraordinary technology will have a golden opportunity, while those that treat this moment as a headcount problem will shrink to fit it." Shrink to fit is a precise phrase. It names the failure mode without softening it.

The structural question this raises for HR and operations leaders is not "how many roles can AI replace?" It's "what does the work actually require now, and who — or what — should own each piece?" That is a design problem, not a procurement problem. The CHRO of a major Thai bank made this distinction explicit in a recent interview: HR Executive (2025-07-09) frames AI integration as fundamentally about redesigning work around people, not retrofitting people around tools. The one capability he insists his own HR team must never lose: genuine human judgment in ambiguous situations. No prompt engineering replaces that.

Atlassian has gone further, creating an entirely new HR role whose explicit mandate is deciding how work splits between humans and AI agents — not just for today's task list, but as a capacity-planning function. HR Executive (2025-07-10) reports that the company also rejects AI mandates — a pointed stance in an industry full of top-down "AI-first" decrees that generate compliance theater rather than genuine adoption. The bet is that voluntary, well-designed integration outperforms mandated usage. The early evidence on botsitting hours suggests that mandate-heavy cultures may be generating exactly the kind of passive, supervisory AI use that produces overhead rather than output.

There's a parallel insight coming from software engineering that applies more broadly than the tech industry. O'Reilly's research makes the provocative claim that O'Reilly (2025-06-30) coding was never actually the bottleneck in software development — which means AI that speeds up code generation may be solving the wrong constraint entirely. The real bottleneck was always scoping, decision-making, and institutional knowledge. Substitute "coding" for whatever your organization's AI deployment targets — report generation, contract drafting, customer triage — and ask whether the same logic applies. Faster execution of the wrong task is still the wrong task.

The organizations that will come out of this period with structural advantage are not the ones that moved fastest on AI tooling. They are the ones that asked the harder prior question: what is the work we actually need to do, and what combination of human and machine capacity accomplishes it with the least waste and the most durability? Headcount reduction is a financial outcome, not a strategy. The companies modeling it as a strategy are confusing a lever for a plan.

If your AI implementation plan looks more like a workforce reduction model than a work redesign model, you have roughly 18 months before that distinction becomes painfully visible in your revenue line.