The most interesting signal in Lenny Rachitsky and Noam Segal’s 2026 survey of tech workers is not that people are anxious about AI, rather that AI is becoming an identity filter.
Some workers feel amplified. Others feel destabilised, diminished, or quietly exhausted. The survey captures a workforce that is more productive and more burned out; more technically capable and less certain of its value.
That is the real leadership problem.
AI is not only changing how much work people can produce. It is changing what organisations expect from them before roles, quality standards, management systems, and career paths have been redesigned around it.
Focus On: The New Performance Baseline
Once AI makes higher output possible, higher output quickly becomes normal.
This is the hidden mechanism behind much of the current workplace tension. The time saved by AI is rarely returned as slack, learning, judgement, or deeper craft. It is absorbed into the operating system as a new expectation.
A marketer who can produce five campaign variants is soon expected to produce ten. A product manager who can draft a sharper PRD in half the time is soon expected to carry more scope. A designer who can explore more visual options is judged against a wider field of possible outputs.
The paradox is now visible in the data. Workers say AI makes them faster, but speed is not translating cleanly into organisational performance. Glean’s 2026 Work AI Index found that workers report significant time savings, but much of the gain is consumed by “botsitting”: feeding context, checking outputs, debugging mistakes, and cleaning up work.
Workday’s 2026 research points to the same issue from another angle: employees may save time with AI, but a large share of that saving is lost to rework, verification, rewriting, and correction.
This matters because many organisations are still measuring AI adoption as activity: seats, usage, prompts, licences, tool penetration.
They should be measuring net value. It is the difference between embedding AI and merely bolting it on.
The Workforce Is Splitting
The Lenny and Segal survey is useful because it moves beyond generic sentiment. It shows a split between those who feel amplified by AI and those who feel destabilised or diminished by it.
That split is all about agency.
The amplified worker has found a way to preserve judgement while increasing leverage. They use AI to accelerate research, sharpen drafts, explore options, reduce friction, and compound their own taste.
The destabilised worker experiences something different. AI makes the work faster, but also less legible. Their sense of craft weakens. The bar moves without being named. Their value becomes harder to explain.
This is where leadership teams often misread adoption.
They assume AI maturity is a skills problem. Train people, give them tools, set targets, and usage will rise. Some of that is true. But the deeper issue is role identity.
People do not resist AI only because they lack training. They resist when AI changes the psychological contract of their job faster than the organisation changes the job itself.
The Hidden Labour Tax
The productivity story is also being distorted by hidden labour.
AI creates outputs quickly. It does not automatically create trustworthy work.
Someone still has to check the argument, clean the prose, correct the analysis, validate the source, test the spreadsheet, challenge the assumption, remove the hallucination, and decide whether the output is good enough to carry the company’s name.
That work is often invisible.
It does not appear in the automation case. It is not counted in prompt-volume dashboards. It is rarely discussed in executive updates that celebrate time saved.
But it is where much of the real cost sits.
I have named this before as the AI dividend trap: organisations overestimate the return because they count first drafts and ignore review burden. They celebrate output volume while quality assurance shifts quietly onto already stretched employees.
This is particularly acute in knowledge work, where the cost of a bad output is often delayed. A weak customer insight, flawed market analysis, generic campaign idea, or misleading executive summary may not fail immediately. It simply lowers the quality of decisions.
Craft Roles Are Exposed First
The anxiety among designers, researchers, writers, strategists, and other craft-heavy roles deserves attention.
These roles are not exposed only because AI can generate artefacts, but also because artefacts are easier to see than the judgement behind them.
A design mock-up is visible. Taste is harder to measure.
A customer interview summary is visible. The researcher’s sense of what was unsaid is harder to quantify.
A campaign draft is visible. The strategic judgement behind the positioning is less obvious.
Nielsen Norman Group’s State of UX 2026 makes this point in a different way: the replacement narrative around UX has been misleading, but it has become convenient in an environment of layoffs, hiring freezes, and pressure to prove business impact.
That is the risk for many craft roles. AI does not need to replace them completely to weaken them. It only needs to make their visible outputs look cheaper while their invisible judgement remains underpriced.
Leaders should be very, very careful. If they automate around craft too aggressively, they may damage the very capabilities AI does not handle well: taste, synthesis, customer empathy, ambiguity tolerance, and strategic restraint.
Strategic Implications for Leaders
Measure workload, not just output. If AI saves three hours and creates two hours of checking, correction, and coordination, the gain is smaller than the dashboard suggests. Track review burden, rework, error rates, decision quality, and time-to-approved-output.
Redesign roles before raising expectations. Do not hold people to AI-enabled performance standards while leaving their role definitions unchanged. If the baseline has moved, say so. Then change scope, incentives, training, and evaluation criteria accordingly.
Protect judgement as a scarce capability. The goal is not more artefacts. It is better decisions. Leaders should distinguish between work that can be accelerated, work that must be reviewed, and work where human judgement remains the source of value.
Make managers the sense-making layer. Managers now have to explain what good looks like in an AI-assisted workflow. They need to set the quality bar, allocate capacity, prevent hidden overload, and help people understand where their value sits as the work changes. As I argued in The Middle Management Problem, agentic AI does not remove the coordination layer; it exposes how badly it needs redesigning.
Treat AI adoption as a talent segmentation risk. The amplified group will pull ahead quickly. The destabilised group may disengage, perform defensively, or leave. Both outcomes require active management. AI capability is becoming part of talent strategy.
The Real Test
The next phase of AI adoption will be won by organisations that redesign work around the right outputs and that means being honest about the trade-off. AI can increase capacity. It can also raise expectations, compress quality, hide labour, and destabilise professional identity.
The leadership imperative is to measure whether the organisation is becoming sharper, healthier, and more capable as a result.
Follow me
That’s all for this week. To keep up with the latest in generative AI and its relevance to your digital transformation programs, follow me on LinkedIn or subscribe to this newsletter.
Disclaimer: The views and opinions expressed in Chronicles of Change and on my social media accounts are my own and do not necessarily reflect the official policy or position of S&P Global.