Trust But Verify, One Level Up
Last week I wrote about botsh*tting — Glean’s term for shipping AI-generated work you haven’t verified, don’t fully understand, and couldn’t defend if asked. Nearly seven in ten AI users admit to it. Remove the incentive to check the work, and eventually people stop checking it.
That’s the individual-contributor version of a problem that doesn’t stay there. This week, one level up: leaders are running the same play, with more at stake when it’s wrong.
Klarna thought AI could take over customer service from its human reps. It built an AI assistant with OpenAI, cut roughly 700 support roles, and handed it three-quarters of customer interactions. Within a year, it was hiring people back. CEO Sebastian Siemiatkowski admitted cost had become too dominant a factor in a decision meant to be about quality. Quality was what dropped. The AI handled routine queries fine, not the ambiguous, judgment-heavy calls that used to go to a person with context to act under pressure. Different mechanism than an entry-level pipeline problem, same root failure: remove the judgment a system depends on before confirming it can run without it. The system doesn’t fill that gap. It exposes it.
Klarna isn’t an outlier — it’s one entry in a trend that got a name this year: boomerang hiring. Robert Half found 29 percent of companies that cut roles for AI have already rehired into them. Fifty-five percent of those executives now call it a mistake.
I’ve spent my last several posts pulling one thread: we say we want critical thinkers, never defined what that means, and we’re quietly eliminating the entry-level roles where the skill develops. Klarna isn’t that story — the cut was wholesale, not entry-level. But it’s the same failure at a different altitude: leaders shipping a decision they hadn’t done the thinking to defend, the way an individual contributor ships AI output they haven’t verified. Last week’s version had a person attached. This week’s has a balance sheet.
Now look at IKEA. Same technology, same cost pressure, same customer-service function Klarna just relearned the hard way. When AI took over nearly half its customer service inquiries, leadership didn’t lay off staff — it reskilled 8,500 call-center workers into remote interior design consultants and built a profitable service team out of the people other companies were cutting.
Two organizations, two answers. One lurched and paid to undo it. One gave direction.
Judgment about context separated them — what the technology replaces, what it depends on, what your people can become if you redirect instead of remove them. Not intelligence. Not budget. Not the AI.
Josh Bersin’s had a name for this for years: the T-shaped professional, deep expertise in one domain paired with working knowledge across many. AI is commoditizing the vertical bar — deep domain knowledge is a prompt away for anyone now. The horizontal bar still has to be earned, one context at a time.
Which makes me wonder about my own generation. GenX has absorbed an entire working life of technology transitions — mainframe to PC, dial-up to broadband, desktop to cloud, now AI. That’s not nostalgia. It’s a pattern library, and my own path through it was anything but straight. The wandering built the judgment.
Skip the technology roadmap. The real audit is about people: how many of the people making your AI workforce decisions have led through more than one technology transition, worked more than one function or industry, watched a “this changes everything” wave up close and seen what actually changed? If the answer is almost none, the risk isn’t moving too slowly. It’s lurching.
Reagan called it “trust but verify.” Last week that was the person checking the AI’s output. This week it’s whether leaders verify their own workforce bets before they ship them — same discipline, aimed one level up.
AI can generate options all day. Knowing which one fits still belongs to a person — and it goes to whoever carries the widest “T” into the room.