AI Didn't Take the Jobs. It Split the Workforce, and Nobody Governs the Breaking Half

Ask 5,920 tech professionals how AI has changed their work and they do not answer as one workforce. They answer as four. In the second annual survey from Noam Segal and Lenny Rachitsky, published this month, respondents sort themselves by AI identity: 49% say AI has amplified them, 27% say it redefined their role, 14% say it destabilized them, and 5% say it diminished them. That self-assigned identity, according to the data, predicts job satisfaction better than role, seniority, or company size. The person sitting next to you may be living in a different labor market, and your org chart cannot see the line between you.

The Split Is the Unit Now

For two years the workforce debate ran on a single axis: will AI replace the job or not. The survey moves the interesting variable somewhere else. Only 22% of respondents report fear of job loss. The dominant experiences are not extinction, they are amplification and destabilization happening inside the same team, the same title, the same pay band.

The amplified half, at 49%, describes leverage. AI removes the parts of the work they never valued and lets them operate at a level their seniority alone would not reach. The destabilized and diminished, at roughly 19% combined, describe the opposite. The ground under their expertise moved, the skills that defined their value are being commoditized, and they are running to stay in place. Both groups are reacting to the same tool. Access is identical on both sides. Identity is what diverges.

This is why the split is the governance unit. A workforce policy written for the average employee now governs nobody, because the average hides a population that is thriving and a population that is quietly coming apart. We argued in The Two-Clock CEO that scale-stage leadership now runs two operating cadences at once. The people-governance version of that problem is sharper. Two workforces, one payroll, and the metrics that would tell them apart do not exist in most companies.

The Squeeze, Not the Robot

The clearest finding in the data is where the pain actually comes from. Respondents report a productivity surge: 82% say AI has made them more productive. In a healthy system that surplus flows somewhere visible, into shorter weeks, higher output priced accordingly, or slack for deeper work. In this data it disappears. 51% report the fear of more work for the same pay, and only 22% fear losing the job at all. The threat people name is the treadmill speeding up while the paycheck holds still.

The productivity gets silently reabsorbed as higher baseline expectations. What took a week is now expected in a day, so the day fills with five times the work, and the gain never shows up as relief. Meanwhile 41% of respondents worry that quality is declining even as speed rises. That last number is self-reported sentiment, not an audited defect rate, and it should be read as what practitioners believe is happening to their craft. Belief still matters here, because the people closest to the output are the early sensor for quality debt, and right now the sensor is flashing while no dashboard records it.

Put the three numbers together. Output up 82%, quality worry at 41%, and the surplus captured as expectation rather than value returned. That is the shape of a squeeze, and it is invisible to any leader watching only velocity. We described the measurement side of this in The AI Workforce Inflection: when the only thing you count is speed, you optimize for the number that hides the cost.

The Morale Bill Is Coming Due

The sentiment data attached to the split is not subtle. Burnout among respondents reached 55.7%, up roughly 11 points from 44.7% the year before. The field’s Net Promoter Score sits at -39. A majority, 53%, say they would now discourage a newcomer from entering the profession. These are people who are, on average, more productive than they have ever been, reporting that they would tell their younger self to pick a different field.

A workforce can absorb a hard year. What the numbers describe runs deeper: a structural condition being read as a personal failing by the people inside it. The destabilized 19% are not underperforming. They are receiving a role redefinition with no map, no retraining path, and no acknowledgment that the change is real rather than a skills problem they should have solved on their own. Left unmanaged, that population does not announce itself. It shows up later as attrition, as disengagement, and as the quiet quality erosion the 41% are already naming.

Manager Quality Is the Most Under-Invested Lever

Here the survey points to the one thing leadership actually controls. Only 25.5% of managers were rated effective by their reports. And the payoff for the other side is large: respondents with a manager they rate highly report roughly 65% higher enjoyment of their work. In a moment when the workforce is splitting by identity and the split is invisible in the aggregate, the manager is the only sensor positioned close enough to see it per person.

The manager is who knows that the senior engineer two desks down went from amplified to destabilized when the team adopted a new agent. The manager is who can route retraining to the person losing ground before that person becomes a resignation. That work is not being done, because most companies invested in AI tooling and left manager capability flat. Three-quarters of managers rated ineffective is the failure of the exact layer that a bifurcated workforce depends on.

For a CHRO or a head of people, this reframes the AI budget. The tooling line is funded. The layer that determines whether the tooling amplifies or destabilizes your people is the manager, and it is the most under-invested lever on the board.

Do This Now

Measure the split before you manage it. Add one dimension to your next engagement survey: ask each employee whether AI has amplified, redefined, destabilized, or diminished their work, and read the result by team and by manager, never only in aggregate. That single question turns an invisible fracture into a map. Then fund the manager layer against it, because the manager is the only mechanism that can catch a person sliding from amplified to destabilized while there is still time to route them somewhere better. The companies that treat a splitting workforce as a people-governance discipline, with its own metric and its own owner, will keep the amplified half and recover the breaking one. The companies that keep watching the average will lose both, one resignation at a time.

This analysis synthesizes How Tech Workers Are Feeling in 2026: A Workforce Splitting in Two (Lenny’s Newsletter, July 2026).

Victorino Group helps leaders build the people-governance layer for a workforce where AI amplifies some and destabilizes others. Let’s talk.