There are two kinds of leaders in the mid-market right now, and they believe they have nothing in common.
The first froze entry-level hiring eighteen months ago. AI can do that work now, he says. He has not run a pilot; he has not measured a single workflow. He read the headlines, watched a demo, and stopped backfilling attrition. The savings hit the P&L immediately, and the board was pleased.
The second has held AI at arm’s length. She has seen the hype cycles before; she is not going to bet her operation on a technology that hallucinates. Her competitors can chase the shiny object. She will wait for the dust to settle. She has not run a pilot either.
Each of these leaders looks at the other and sees a cautionary tale. The slasher sees a dinosaur; the skeptic sees a gambler. What neither sees is the mirror.
The inkblot doesn’t care what you see
Here is the uncomfortable truth: both leaders reached their conclusion the same way, which is to say, without evidence. Neither tested the technology against their actual work. Neither measured anything. Both looked at AI and saw exactly what they were predisposed to see, and did not notice that it was materially redrawing the landscape of their sector.
That is not a technology assessment. That is presumption; and presumption pointed in opposite directions is still the same act. It is also, in the clinical sense, a Rorschach test.
Show an ambiguous inkblot to a hundred executives and you learn nothing about the inkblot; you learn a great deal about the executives. AI is functioning the same way in boardrooms today. The cost-cutter looks at AI and sees a permission structure for the headcount reduction he already wanted. The skeptic looks at the same technology and sees confirmation that patience was the right posture all along. The variable was never the evidence. It was what each leader wanted to believe before AI ever arrived.
The data behind the faith-based freeze
The scale of untested conviction is now measurable. A Harvard working paper analyzing 66 million workers across 280,000 firms found that companies adopting generative AI cut entry-level employment roughly 9% within six quarters, while senior headcount at the same firms kept growing. The most revealing detail: separations actually fell at these companies. Nobody was fired and replaced by a validated AI workflow. Leaders simply stopped backfilling attrition and assumed the technology would absorb the work.
Assumed. Not confirmed.
The researchers traced the timing, and it tells the story plainly: hiring pullbacks began almost immediately after AI entered the corporate conversation, well before the automation gains materialized. Companies were adjusting for capabilities they anticipated rather than capabilities they had proven.
Now set that against a second data point. A 2026 survey of 500 mid-market decision makers found that 91% expressed confidence in their internal AI expertise, yet only 10% had successfully scaled all of their AI initiatives beyond pilot stage. Read those numbers together and the picture sharpens: the same organizations that cannot get a proof of concept into production are confident enough to restructure their talent pipeline around AI’s presumed capabilities.
Confidence in AI strategy has become completely decoupled from competence in AI execution. That gap is where careers, pipelines, and eventually companies quietly break.
To the reluctant leader: this is about you too
If you are in the skeptic’s chair, you have probably read the foregoing with some satisfaction. Those reckless cutters; you always knew.
Hold that thought, because the next part is for you.
Your caution rests on exactly as much evidence as their naive “magic wand” optimism. You have not measured which of your workflows AI handles well, so your restraint is not prudence; it is a guess wearing prudence’s clothes. The slasher guessed high, and you guessed low, but you are both guessing. The inkblot showed him opportunity and showed you risk, and neither of you checked.
The cost of presumption is simply distributed differently. His error compounds invisibly: work quietly piles onto senior staff, quality erodes at the edges, and the bench gap surfaces in three to five years when there is no one ready to promote. Your error compounds invisibly too: every quarter of untested waiting, a competitor validates one more workflow, banks one more margin point, and learns one more organizational lesson you have deferred. His bill arrives as a talent crisis; yours arrives as a competitiveness crisis. Both bills arrive.
And there is a second-order cost coming that will land on your desk specifically. The faith-based cutters are going to generate spectacular failure stories over the next two years. “We tried AI, cut the team, and it didn’t work” will echo through every industry conference, and it will feel like vindication for the cautious. It will not be. What those companies tried was a hiring freeze plus optimism; the technology was never actually deployed against the work. If you let those stories harden your skepticism, you will have learned the wrong lesson from someone else’s unforced error.
The only posture that isn’t projection
There is a third kind of leader, still rare, who treats AI as neither savior nor mirage but as a claim to be tested.
The discipline is not complicated:
Instrument before you conclude. Pick the workflows where AI’s impact would matter most, define what success looks like in measurable terms, and run structured pilots against real work with real stakes. Not a demo; a deployment with a scoreboard.
Restructure only what you’ve validated. If a pilot proves AI absorbs 60% of a role’s task load, you have earned the right to redesign that role. If you haven’t run the pilot, you haven’t earned anything; you are just redecorating your prior beliefs with an org chart.
Protect the pipeline while you test. Junior talent is where your future senior bench comes from, and it is also where AI fluency enters your organization fastest. Cutting it on faith is borrowing against a future you will still have to live in.
Separate the technology’s capability from your organization’s readiness. Most stalled AI initiatives fail on data quality, process clarity, and governance, not on model performance. Knowing which problem you actually have is itself a finding that only testing reveals.
None of this is exotic. It is the same evidence-based leadership you already apply to acquisitions, capital projects, and market entry. The strange thing about this moment is how many otherwise rigorous executives suspended that standard the moment the technology got interesting.
The inkblot will keep showing you whatever you brought to it. The only way out of the Rorschach test is to stop presuming and start exploring; the leaders who do will spend the next five years compounding advantages while the believers and the doubters split the cost of presumption between them and argue about what the picture means.
