What made Maven different wasn’t the technology. It was the decision to treat AI not as an experiment to be managed, but as an organizational transformation to be owned. Senior leaders fought for it personally and bureaucratically. Workflows were dismantled, not augmented. Outcomes, what it enabled warfighters to actually do, were the only measure that counted. That discipline is the reason it worked.

I tell that story because Stanford’s 2026 AI Index, released in April, confirmed something corporate America has been quietly conceding on earnings calls all spring: the country building the world’s most powerful AI ranks 24th in using it. American adoption sits at 28.3 percent. Singapore is at 61. The UAE is at 54. Goldman Sachs noted that AI investment contributed “basically zero” to U.S. GDP growth last year. America is falling behind not because of its models or its chips. It is falling behind for the same reasons the Pentagon almost lost Maven, and the fix is the same one that saved it.

America is building the world’s most powerful artificial intelligence but stumbling badly in applying it. For now, China’s edge isn’t superior technology. It’s superior integration. And in the race that matters most, integration is what wins.

Beijing’s “AI Plus” initiative is explicitly aimed at embedding AI across manufacturing, logistics, scientific research, health care, education, and government operations. In manufacturing in particular, the emphasis goes beyond generic assistants toward sector-specific models, industrial datasets, intelligent agents, and large-scale workflow integration. China is not treating AI as something to build. It is treating it as operating infrastructure. It is not debating whether AI can be controlled or contained. It is deploying it.

History is full of countries and companies that led in invention and failed in implementation. We rest comfortably on our national AI model and chip prowess, on leaderboard rankings and benchmark scores, but those achievements, while herculean, are not what ultimately determines who wins.

This is why so much of the corporate AI conversation feels wrong. Executives talk spending and use cases. Procurement is easy, use cases are the wrong measure, and reorganization is threatening. It forces leaders to confront the questions they spend most of their time avoiding: Which decisions can be automated? Which reviews can be eliminated? Which workflows should disappear?

Most companies do not want answers to those questions. They want the optics of innovation without the pain of change.

Building an organization around AI is hard. And that is precisely the challenge American enterprise is failing to confront. The companies that survive the next decade will be the ones that take themselves down to the studs, dismantling legacy processes, legacy org charts, legacy assumptions about how work gets done, and rebuilding as AI-native. Not AI-augmented. AI-native. There is a difference, and it is the difference between competitiveness and obsolescence.

This isn’t a theoretical risk. It is an emerging crisis with a specific shape: a white-collar reckoning likely worse than the blue-collar offshoring wave of the early 1970s, but faster and less forgiving. If American companies do not move aggressively, they will face an accounting from competitors in the East who are leaner, faster, and unencumbered by the institutional inertia that is fast becoming corporate America’s most expensive liability.

Maven was never a pilot program. It was an organizational transformation, one that proved a premise many in the building doubted: that commercial software and AI could be embedded into the workflows of the world’s largest bureaucracy, on some of the most complex tasks, and produce results at a scale legacy systems could never achieve.

The Defense Department is not exactly known for its procurement successes. When it comes to software, it is known for producing expensive, overdue solutions that fall short. And yet on Maven, it got the fundamental question right, not by launching a technology experiment, but by treating AI as institutional change that demanded executive ownership from the start. That is the lesson corporate America must embrace.

Most large companies today are running AI experiments. Few are running AI transformations. The difference is not technical. It is organizational. And it comes down to three failures that Maven’s experience exposes with uncomfortable clarity.

First, AI at scale requires executive ownership, not delegation. At the Pentagon, Maven worked because senior leaders owned it personally and fought for it bureaucratically. In too many corporations, AI strategy is delegated to a chief AI officer or an innovation lab, organizational constructs designed to signal progress without disturbing the existing power structure. That is a recipe for pilot purgatory.

Second, AI requires dismantling legacy processes, not layering on top of them. Companies that bolt AI onto existing workflows get marginal efficiency gains and call it transformation. If your org chart, approval chains, and operating cadence look the same as they did before your AI initiative launched, you have not transformed anything.

Third, AI demands delivery accountability. Maven was measured by operational impact, what it enabled warfighters to do that they could not do before. Too many corporate AI programs are measured by activity: models trained, proofs of concept completed, partnerships announced. Activity without outcomes is theater.

The answer to China’s AI advance is not that we have better models or better processors. It is execution. America can still lead the AI era. But it will not do so by resting on leaderboard rankings or fretting about who has what chips. It will do so only when its institutions look internally and prove they can do something harder: change.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.