Jesse Jacoby
Two directors in the same function at a client of mine, same leader, same AI tools, described the same eighteen months in language so different you would not believe they worked at the same company. The first told me AI had been the most energizing development of her career. She was doing analysis she never had the hours for and talked about her job the way people talk about a new hobby. The second told me he no longer knew what he was for. His judgment used to be the product. Now the model produced a serviceable version of it in ninety seconds, and he spent his days checking work he used to do, unsure whether he was the expert or the reviewer.
Both were rated strong performers, and neither was wrong about their own experience. The organization had one AI strategy, one training program, and one internal narrative about augmentation and opportunity. What it did not have was any recognition that all of it was landing on two completely different workforces.
The Question That Predicts Everything
The second annual tech worker sentiment survey from Noam Segal and Lenny Rachitsky, published this month, put hard numbers to something I have been watching form in client organizations for two years. Significant burnout jumped from 44.7 percent to 55.7 percent in a single year. Career optimism fell from 54.8 percent to 48.7 percent.1 But the finding that should stop executives cold is not either of those numbers. It is what best predicts them.
The researchers asked how working with AI had shifted people’s sense of themselves as professionals. Roughly half said amplified. About a quarter said redefined. Nearly one in five said destabilized or diminished. That single answer predicted career optimism more powerfully than role, seniority, and company size combined, and the gap between the amplified and the diminished was the largest effect in the dataset by a wide margin, roughly three times the size of the well-documented advantage founders hold over everyone else.2
The sample is tech professionals, so hold the specific figures loosely. The pattern is what matters. The most consequential segmentation inside a knowledge workforce today is not function, level, tenure, or location. It is what AI has done to a person’s sense of their own value, and almost no organization is measuring it.
The Fear Is Not Replacement. It Is the Ratchet.
Every board conversation I sit in about AI and the workforce frames the human risk as displacement. The workforce is worried about something else entirely. Only 22 percent named losing their job to AI as a top concern. Far more named being expected to do more for the same pay (51 percent), getting locked into an unsustainable pace (46 percent), and watching the quality of their own work decline (41 percent).3
This is the productivity ratchet, and it is the defining organizational dynamic of the moment. AI creates real capacity. The capacity gets absorbed into higher expectations. The new output level becomes the baseline, the baseline resets the following quarter, and nobody ever experiences the gain as relief. Eighty-two percent said AI is making them measurably more productive, and burnout climbed eleven points in the same year.4 Those two facts are not in tension. The second is a consequence of how organizations chose to spend the first.
Leaders keep waiting for the productivity dividend to show up in engagement scores. It will not, because the dividend was spent before anyone decided how to spend it. Efficiency gains never return themselves to people.
Smiling Exhaustion
The divide is easy to miss because it does not present as unhappiness. Seventy-seven percent held at least one positive and one negative feeling about AI at once, and the average person selected more than five emotions. Curious and overwhelmed. Excited and tired. Nikhyl Singhal called it smiling exhaustion, and the phrase is exact.5
This is why your engagement survey will not find the fault line. Ambivalence averages out. A workforce split evenly between energized and depleted produces the same aggregate score as one that is uniformly fine, and the instrument reports stability while two very different populations drift apart underneath it. The most telling number may be this one: 53 percent would steer a newcomer away from their own field, and a third of those who call themselves optimistic about their own careers would still do it.6 People have made peace with their own trajectory while quietly concluding the path behind them no longer works. That does not register as disengagement. It shows up two years later as an empty pipeline.
The Divide Is Probably Manufactured, Not Inherited
Here is the part leaders can actually control. The split between amplified and diminished is not a fixed trait of individuals, and the most plausible explanation for it is not temperament. It is whether the organization redesigned the work or simply handed out the tools.
Deloitte’s 2026 Global Human Capital Trends research found that only 6 percent of leaders say they are making real progress designing human and AI interactions. Sixty-six percent of C-suite leaders say their traditional functions have to change, while 7 percent report progress toward changing them.7 Organizations have invested heavily in access and fluency, and almost not at all in redesigning roles, workflows, decision rights, and career paths.
When the organization does not redesign the work, every individual has to redesign it themselves, alone, in the middle of their day job. Some find that exhilarating. Many find it destabilizing. On that reading, the divide is less a story about who is adaptable and more a story about what leadership left undone, and who had the confidence, autonomy, and slack to cover the gap. One suggestive data point: people at companies of 10,000 or more were the most likely to feel destabilized, people at the smallest companies the least.8 Scale does not make people less capable. It makes self-redesign nearly impossible.
The Manager Is the Variable
The most actionable finding in the dataset is also the oldest one. Manager effectiveness remains the single strongest driver of burnout, ahead of role, company size, and AI sentiment. Workers with an extremely effective manager report roughly 65 percent higher job enjoyment. And yet only about a quarter rate their manager as highly effective, while more than a third rate theirs as ineffective, numbers that have barely moved in a year.9
This matters more in an AI transition than it did before, because the manager is the only person positioned to do the things that decide which side of the divide someone lands on. They can reset expectations when the ratchet tightens. They can renegotiate what the work is rather than just how fast it gets done. And they can tell a person what they are for when the model has taken over the task that used to answer that question. No enterprise communication does any of that. It happens in one-on-ones or it does not happen. Fund AI capability without funding the managers who have to absorb it, and you have bought the tool while defunding the thing that decides whether it creates value or casualties.
What to Actually Do
Measure the divide directly. Add the identity question to your next pulse survey, then segment every other result by the answer. You will learn more from that one question than from ten more items about tool satisfaction, and you will finally see the population your averages are hiding.
Decide how the capacity gets spent. If AI frees twenty percent of a team’s time and nobody makes an explicit decision, the answer defaults to more output. Make it a governed choice: some to throughput, some to quality, some returned as genuine capacity. A leadership team that cannot say out loud how it is allocating its productivity gains has already given all of it to the ratchet.
Redesign the work, not just the toolkit. Training tells people what the tool can do. It does not tell them what their job is now. Role redesign, revised decision rights, updated definitions of good work, and honest career path conversations are the actual intervention. This is the 6 percent problem, and it is where the competitive separation happens.
Equip managers first. Before the next wave of licenses, give managers language and permission for the conversation about what a person is for now. That conversation is the difference between amplified and diminished, and most managers have never been asked to have it.
One Strategy, Two Realities
Go back to those two directors. Same company, same tools, same training, opposite experiences. The organization was not running one AI transformation. It was running two, and only one of them was on the roadmap.
The ground is moving under every knowledge worker in your organization right now, and some experience that as a launch while others experience it as an earthquake. Which one a person gets is not fate. It turns on whether the work was redesigned around them or dropped on them, and whether anyone with authority helped them answer the question the technology raised about their own value. That is a leadership choice, and in most organizations it is being made by default. So the question worth putting to your leadership team this quarter is a simple one: do you know which of your people feel amplified and which feel diminished, and are you prepared to lead as though the answer matters?
References
- Significant burnout rose from 44.7% to 55.7% year over year; career optimism fell from 54.8% to 48.7%. Noam Segal and Lenny Rachitsky, “How tech workers are feeling in 2026: a workforce splitting in two,” Lenny’s Newsletter, July 7, 2026.
- Respondents described their professional identity under AI as amplified (49.0%), redefined (27.4%), destabilized (13.9%), diminished (5.0%), or unchanged (3.2%). AI-identity stance was the strongest predictor of career optimism in regression, ahead of role, level, and company size combined. The amplified-to-diminished gap measured Cohen’s d ≈ 1.55, against d ≈ 0.56 for the founder effect. Segal and Rachitsky, Lenny’s Newsletter, July 7, 2026.
- Top career concerns about AI: expected to do more for the same compensation (51%), trapped in an unsustainable pace (46%), quality of work declining (41%), losing my job to AI (22%). Segal and Rachitsky, Lenny’s Newsletter, July 7, 2026.
- 82% reported AI made them at least moderately better at their job; 49.4% said “very much” or “extremely.” Segal and Rachitsky, Lenny’s Newsletter, July 7, 2026.
- 77% selected at least one positive and one negative emotion about AI; the average respondent selected more than five. The phrase “smiling exhaustion” is Nikhyl Singhal’s, quoted in Segal and Rachitsky, Lenny’s Newsletter, July 7, 2026.
- 53% would not recommend a career in their role to someone starting out, an NPS of −39; roughly a third of self-described optimists would still steer a newcomer away. Segal and Rachitsky, Lenny’s Newsletter, July 7, 2026.
- 6% of leaders report progress designing human-AI interactions; 66% of C-suite leaders say traditional functions must change while 7% report progress; 65% say culture needs to change significantly because of AI. Deloitte, 2026 Global Human Capital Trends.
- 23% of respondents at organizations of 10,000 or more reported feeling destabilized, against 15% at organizations of 1 to 10. Segal and Rachitsky, Lenny’s Newsletter, July 7, 2026.
- Manager effectiveness was the strongest driver of burnout in the dataset, ahead of role, company size, and AI sentiment. Respondents with an extremely effective manager reported roughly 65% higher job enjoyment. 25.5% rated their manager highly effective; 36.5% rated theirs ineffective. Segal and Rachitsky, Lenny’s Newsletter, July 7, 2026.
Survey findings reflect a self-selected sample of technology professionals and should be read as directional for other sectors.
