The press releases all rhyme by now. There is the foundation-model partnership. There is the chief executive describing a transformative reinvention of the roadmap. There is the small chatbot in the lower right corner of the support page. There is the summarization feature in the dashboard nobody actually opens. The company, by the company’s own account, is now an AI company. The architecture beneath the announcement, on inspection, has not been touched.
Inside several of the larger software firms the cycle is producing a particular kind of frustration, and the frustration has begun to organize itself into an argument. The argument is narrower than the standard critique of AI hype, and harder to wave off. Most of what is currently being announced as AI strategy looks like a thin layer painted onto an unchanged product. The data model is the same. The workflows the customer actually depends on look the way they did eighteen months ago. The marketing has changed. Almost nothing else has.
Pearce Dolan, head of product at Deel, the global employment platform that crossed a billion dollars in annual recurring revenue last year, has begun making the argument with more directness than most of his peers in the category. He calls it AI theater.
“Most companies doing AI are actually doing AI theater,” Dolan said. “They’ve added a chatbot, bolted a summarization feature onto their dashboard, and issued a press release. What they have not done is embed AI into the actual structure of how they build, operate, and deliver value.”
The framework Dolan offers separates two postures the industry tends to fold together. The first is AI as a layer. The layer sits on top of a product designed for an earlier paradigm. The AI feature is, by his account, a wrapper, sometimes a useful one, occasionally a delightful one, but structurally inert. The second is AI as architecture. An architectural deployment changes the assumptions the product is built on, the way data moves through it, the way work gets done inside the company that ships it.
“There is a critical difference between adding AI as a layer and embedding AI as an architecture,” Dolan said. “A layer sits on top of what already exists. An architecture changes how the thing is built, operated and experienced at its foundation.”
The framework would be merely descriptive if Dolan were not also unusually qualified to make the case. He runs a 200-person product, design and engineering organization at a firm that operates payroll, employment and compliance infrastructure across more than 150 countries. He also writes code. By his own description he has made a deliberate practice of building agents personally and shipping AI-driven features into production, on the theory that when the senior person in the room demonstrates what is achievable, the team’s sense of what fits inside a sprint expands.
The conviction that the senior operator in the room sets the team’s sense of the possible is not a new one in software. In the current moment, however, it is structurally consequential. The cost of building a working software prototype has fallen by orders of magnitude in the past eighteen months. The bottleneck has shifted. The constraint, increasingly, is whether the team has internalized the shift, and whether the leadership has demonstrated, by personal practice, that the shift is real.
A second argument runs through Dolan’s position and may matter more than the first. It concerns what AI is doing to the role boundaries that have governed software organizations for the past fifteen years.
“AI is blurring traditional role boundaries in the best possible way,” Dolan said. “Designers can write code. Product managers can prototype. Engineers can iterate on design. The specialist skills remain, but AI gives every team member a functional baseline in adjacent disciplines. Delivery is no longer gated by job titles.”
His framing for the broader moment is historical, and reaches backwards more than forwards.
“First-principles thinking is the only methodology that survives a platform shift,” Dolan said. “Every significant shift, mobile, cloud, now AI, exposes the same fault line. Companies that built on assumptions versus companies that built on reasoning. When the assumptions change, the first group has to start over. The second group adapts.”
The fintech parallel is one Dolan has returned to before. Legacy banks, when the smartphone arrived, responded by adding mobile applications to existing core systems. The mobile application was the layer. Revolut, where Dolan worked between January 2018 and August 2020, asked a different question, which was what financial infrastructure should look like when the customer was a phone-native user with no loyalty to a branch. The architectural answer is the company that exists now. The legacy banks, by and large, are still painting layers.
The HR technology incumbents, in his telling, are repeating the pattern. They are digitizing paper processes. Deel was built around the question of how global employment should work when the workforce no longer sits inside a single country. The AI cycle is the same fault line, surfaced again. The companies that have mistaken a chatbot for a strategy will discover, if they have not already, that the architecture they have been describing in earnings calls is not actually built around the technology they keep claiming to have adopted.
The provocation is quieter than the phrase AI theater suggests. The people most loudly announcing their AI strategies are, by Dolan’s account, the people whose underlying products have not been touched. The tell is whether the team building the system can describe what changed at the foundation. If they cannot, the layer is the strategy. The layer, eventually, is replaced by something built differently.
