![\"TOPSHOT-CORRECTION](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/07\/1785443056_128_0x0.jpg\")
<\/p>\n<\/p>\n
AI can copy the face. Getting it to hold the judgment behind one is the harder problem. (Photo by VALENTIN FLAURAUD\/AFP via Getty Images)<\/p>\n
AFP via Getty Images<\/p>\n
Every time Jeff Eyet finishes a lecture at UC Berkeley\u2019s Haas School of Business, he does something most people would find masochistic. He records his own delivery, feeds the recording and his slides back into a large language model (LLM), and asks it one question: what critique would an MBA student offer?<\/p>\n
“It\u2019s uncomfortable, but it’s the highest-return thing I do,” Eyet, who leads the AI faculty program at Haas, told me. He guesses fewer than one in ten of his colleagues does the same.<\/p>\n
That habit puts him on the right side of a line most companies are on the wrong side of. The Census Bureau’s Business Trends and Outlook Survey<\/a> found that only about 18% of U.S. firms were using AI in any business function by early 2026, with manufacturers trailing at 14%. Read fast, that looks like a simple race between the companies that moved and the ones that didn’t. Eyet thinks the race is being scored wrong. Most of the firms that adopted, he says, bought tools instead of building capability. “Moving early is not the same as moving well, and adoption surveys can’t tell the difference.”<\/p>\n Eyet is not selling AI, which is why founders let their guard down with him. “I don’t teach AI theory. I teach teams to ship real workflows with human judgment at the center,” he said. Right now, he argues, most companies are aiming the technology at the half of the job that matters least.<\/p>\n Why Is Everyone Pointing AI The Wrong Way?<\/p>\n Almost every AI pitch is about producing something. Write the email, draft the deck, generate the campaign. Eyet’s bet cuts the other way. The richest use is to point the model at work you have already finished and tell it to take the work apart.<\/p>\n “Everyone is pointed at generating net-new content. Almost nobody is pointed at critique,” he said. Nobody is selling this, which is exactly why it hides in plain sight. No license, no dashboard, no procurement cycle. You hand the model your earnings script, your board deck, your fundraising pitch, and you tell it to play your sharpest adversary.<\/p>\n What Does It Look Like To Red-Team Yourself?<\/p>\n The practice has a name in his world: red-teaming. Take the number you are about to present, hand it to a model, and give it one instruction. Argue the other side, before the people in the room do it for you.<\/p>\n Finance is where he sees this pulling companies apart. A CFO used to own the narrative in a board meeting. Now half the room has already run the numbers through their own models and arrived with a story. “Your job used to be telling it,” Eyet said. “Your job now is unwinding theirs, and nobody gives you extra time on the agenda for that.”<\/p>\n Some finance chiefs are already rebuilding the seat around that reality. At Hewlett Packard Enterprise, CFO Marie Myers rebuilt her team’s weekly performance review around an AI system her group nicknamed Alfred, cutting the reporting cycle by about 40%<\/a>. The lesson she stresses is not about the software. “Even though you have all these AI capabilities, you actually have to have a human in the loop,” Myers told Fortune. She spent over a year reskilling her finance team to run the tool, not trust it blindly.<\/p>\n