In May 2025, Klarna's chief executive Sebastian Siemiatkowski stood in front of investors and reversed a decision he had spent two years bragging about. The buy-now-pay-later company had replaced the workload of 700 customer service employees with an AI assistant, and Siemiatkowski had turned that number into a talking point — proof that Klarna was the AI-native company its IPO pitch deck claimed to be. Now he was hiring humans back. "We focused too much on efficiency and cost," he told reporters. "The result was lower quality, and that's not sustainable."
Fourteen months later, Mark Zuckerberg told Meta employees something similar, in language considerably less rehearsed. After a restructuring that eliminated roughly 8,000 jobs in the name of an AI-first reorganization, Zuckerberg admitted the company had "miscalculated" the timing. An internal memo from chief technology officer Andrew Bosworth described the rollout of Meta's own Applied AI division in one word: "atrocious."
Two of the most AI-forward companies on earth, a little over a year apart, arrived at the same confession: they had adopted the technology and skipped the part where the organization actually changes. They are not outliers. They are the visible edge of a number that belongs on every board deck and rarely gets there — 80% of companies have deployed AI in at least one core business function. Only one in five have redesigned how the work in that function actually gets done. A separate industry survey puts a finer point on it: 37% of organizations are using AI at a surface level, with little or no change to the underlying process it was purchased to improve.
None of that is a story about the technology underperforming. The tools mostly work. It's a story about what companies do — and mostly don't do — with them once the press release goes out.
Companies Get Two Steps Into an Eight-Step Process, Then Stop
In 1996, the Harvard Business School professor John Kotter published a change-management model that has since become close to gospel in every MBA program and consulting deck on earth. It has eight steps: create urgency, build a coalition, form a vision, enlist people behind that vision, remove the structural barriers in the way, generate short-term wins, sustain momentum, and finally institutionalize the change — bake it into how the company hires, promotes, and measures itself going forward.
Read the AI rollout playbook most companies have actually followed against those eight steps, and the pattern is exact. Step one, urgency, is easy — a memo from the CEO, an earnings-call mention, a competitor's press release. Step two, the coalition, is also easy — stand up an AI task force, appoint a head of AI, buy the enterprise license. Most companies stop right there and call it a transformation. Step three is where it gets expensive, because forming an actual vision means deciding, specifically, which roles get consolidated, which approval chain gets deleted, and whose job changes shape. That is a political decision with a name attached to it. A task force can't make it. Only a leader spending real capital can, and most declined to spend it.
Buying the tool is the urgency memo. Redesigning the org is steps three through eight. Most companies stop at two and call it a transformation.
What the Reversals Actually Reveal
Klarna's AI customer service push is, in retrospect, a textbook stall at step two. The company had genuine urgency — a looming IPO, a story to tell about margins — and a genuine coalition, since an AI vendor relationship is a coalition of exactly two logos. What it never built was step three: a vision for what "AI-native customer service" should actually feel like to a customer who needed a real answer, not a fast one. Siemiatkowski's own postmortem makes the diagnosis for you. "From a brand perspective, a company perspective," he said, walking back the strategy, "I just think it's so critical that you are clear to your customer that there will always be a human if you want."
Meta's version of the same stall runs through steps five and six — removing barriers and generating short-term wins. The company had urgency in abundance and a coalition with essentially unlimited capital. What it skipped was the unglamorous work of proving the new approach at small scale before betting the org chart on it. Bosworth's "atrocious" memo and Zuckerberg's "miscalculated" admission are, read together, an account of a company that reorganized around a vision of AI-native work before that vision had survived contact with a single working team.
The Brand Story Outran the Org Chart
There's a second gap sitting underneath the operational one, and it's a marketing problem as much as a management one. The external story — the earnings-call language, the "AI-first" repositioning, the recruiting page that now says "AI-native" — has been getting rewritten faster than the internal org chart that's supposed to back it up. AI-strategy consulting grew roughly 89% year over year in 2025, as companies rushed to buy the narrative alongside the tooling, even as some of the largest firms selling that narrative quietly cut around 10% of their own analyst-level headcount in the same period. Executive coaching is being repriced by the same logic in the opposite direction: independent research estimates AI can already cover up to 90% of routine day-to-day coaching conversations, at least one major coaching platform cut the price of its AI-delivered tier by roughly 70%, and close to half of Gen Z professionals now say they use AI tools for career decisions that used to go to a manager or a mentor.
Put those two trends next to each other and the shape of the market becomes obvious. The commodity version of expertise — the routine, repeatable, low-stakes advice — is being priced toward zero, fast. Demand is consolidating around the version that comes with a name attached: someone who has actually run the redesign being described, who can be wrong in public, and who is still standing when the plan meets a P&L that doesn't cooperate. Companies are buying that story before they've built the substance behind it. Klarna and Meta are what it looks like when the gap between the story and the substance gets big enough to show up in an earnings call.
What Step Three Would Actually Require
If the goal is genuinely to close the eighty-percent-to-twenty-percent gap — rather than to have a good number for the next board deck — Kotter's model is not subtle about what comes next. A real vision names, specifically, which process gets deleted and which role gets rebuilt, not just which tool gets a seat license. Removing barriers means a leader spending political capital on the manager who liked the old approval chain, not just budget on a vendor contract. Short-term wins have to be measured against the new way of working, not against how many employees clicked "enable" on a new feature. And institutionalizing the change means it eventually shows up in how people are hired, reviewed, and promoted — not just in how they're trained.
Every company on earth can now buy roughly the same AI tool. That was never going to be the differentiator, because everyone in the category has access to it. The differentiator is the unglamorous six-to-eighteen-month stretch of organizational work sitting on the other side of step two — and it is, for most companies, still sitting there untouched.
The tools are installed. The adoption number is on the slide. Almost nobody has read past step two.
