Three senior beauty executives sat on a Miami stage this month and met the question the technology usually dodges: what does your AI do all day? Their answers, reported by Glossy, were chores. The revolution lives in the press release; the work lives in the building.

The distance is the point. In April, Ulta Beauty and Google shared a stage to announce “agentic commerce” — a Gemini assistant drawing on more than 46 million member profiles that could carry a shopper from first question to checkout without the clicks between. Some of the buying would happen inside Google’s own search and app. The language was collapse and transformation. By every measure a keynote is judged on, it was a triumph.

Six weeks later, on Glossy’s summit stage, Josh Friedman, Ulta’s svp of digital and e-commerce, described that system at the desk. The funnel had not collapsed. He was still learning how shoppers use it. The progress he could name was a chatbot fed fresh data: store locations, store hours, the queries the customer-care agent already handles. The most heralded beauty-AI launch of the spring had, by June, learned what time the shop opens.

Tarte was no grander, and the better for it. Jenna Manula Linares, who runs its digital marketing, ends each weekly meeting with a fifteen-minute reckoning: who used AI, and whether it saved them anything. “If it takes you longer than 15 minutes to do something,” she tells the team, “there’s a faster way, and you should learn and try to figure it out via AI.” Creative briefs now demand an AI-written first draft, and a landing-page redesign that once crawled now moves. The gain is real; it buys a faster Tuesday, not a different week.

Beekman 1802 went furthest, and proved the point hardest. Its chief revenue officer, David Baker, built an “agentic staffer” named Zoe to find creators and draft the outreach at scale while the team sleeps. His other initiative was simpler: spot the staff already curious about the tools, and give them room to play. Baker also supplied the obligatory warning: someone who can use AI will come for the job of someone who cannot. The job his own AI came for was writing the emails nobody enjoyed writing.

The budgets follow the stage, not the desk. Boardroom attention and trade press both chase the AI that performs; the value, so far, sits in the AI that files. We have watched this gap before: Glossy’s own leaders’ dinner quietly traded AI deployment for community managers even as the headlines sold transformation. The spectacle is, for now, the part that isn’t real. That is not a failure of the machines — it is a failure of the keynote to resemble the work.

The useful AI and the sellable AI turned out to be different products.

None of this is the executives’ failure; it is, if anything, their candour. Asked to describe the future, they described their Tuesday, and were right to. The stage will keep promising the funnel, because no one ever booked a ballroom to announce that the chatbot has learned the hours.