Why does the same AI tool make one company more productive and another worse off?

This explores why the same AI tools pay off at some companies and stall at others, and how much of that gap comes from management rather than from the technology. The short answer from the corpus is that management decides almost all of it. The surprise is in *how*: good management doesn't skip the early pain of adopting AI. It decides who gets through that pain and comes out ahead.

The clearest evidence comes from US manufacturers. Census-linked data shows that firms adopting industrial AI first get *less* productive, about 1.33% lower at the start, and then grow in the medium term. Those later gains are concentrated in younger, well-managed firms Does industrial AI adoption hurt productivity at first?. This dip-then-rise pattern is called a J-curve. The dip reflects real costs in inventory, robots and staffing, not an accounting quirk. Brynjolfsson, Rock and Syverson explain why the curve exists. AI only pays off after companies make large hidden investments in training, new processes and supporting software, and standard statistics don't count those investments Why does AI productivity appear weak when investment is high?. Management quality, then, is largely the ability to make those investments and survive the dip while they mature.

Why is that investment so hard to get right? Agrawal, Gans and Goldfarb point to a coordination problem. AI makes each person better at their own task, but it also makes their decisions more varied. When decisions depend on each other, that variation causes friction unless the organization communicates and restructures to match Does AI adoption require rethinking how organizations coordinate?. Deloitte sees the same thing with AI agents: projects fail when agents are added on top of old workflows and succeed when processes are redesigned end to end Does agentic AI succeed through process redesign or workflow layering?. Evans adds a less obvious obstacle. Making AI tools easier to build doesn't help much, because most workers don't recognize their own tasks as automatable, and real adoption requires decisions that cut across departments Does easier tool-building actually solve enterprise adoption problems?.

Day-to-day management matters too. Gallup found that employees whose managers actively support AI use report improved workplace culture at nearly twice the rate of others (31% vs. 21%). The data is correlational, though, so good managers and positive teams may simply go together Does manager support actually shape how employees experience AI at work?. OpenAI's enterprise data shows a gap in who actually uses the tools. Marketing staff and early-career workers use ChatGPT far more than executives and senior staff Who adopts enterprise AI first and how do they use it?. The people who would need to redesign the organization are often the ones using AI least.

The advantage also compounds. Firms with more AI exposure replace outside contract labor with AI faster and more cheaply, which suggests that internal AI capability builds on itself instead of spreading evenly Do firms substitute labor for AI at different rates?. As AI agents become more autonomous, management also shifts from incentives to oversight. Humberd and Latham argue that monitoring has to come before any attempt to align an agent's incentives Must AI monitoring precede incentive alignment as agents evolve?. Agency-law scholars warn that the usual governance tools break down when agents make fast decisions nobody can interpret Can traditional governance tools actually control AI agents?. So the management skill that pays off is shifting: from encouraging people to use AI, toward redesigning how work flows and overseeing systems you can't fully read.