How tech workers actually feel about AI in 2026 | Annual AI sentiment survey (Noam Segal)
Lenny Rachitsky (Lenny's Podcast), guest Noam Segal Β· podcast Β· 2026-07-13 Reader π₯οΈ
AI Confidence TheaterBurnout & RecoveryCognitive SurrenderEmotional Impact of AI WorkFounder ModeManaging Up & InfluenceProfessional IdentityPsychological Debt
AI's effect on how tech workers feel about their professional identity β not their role, company, seniority, or any other job characteristic β is now the single largest driver of workplace sentiment, splitting the workforce roughly in half between those who feel amplified and those who feel destabilized or diminished; burnout has surged industry-wide as a result, and effective management is the strongest available lever for countering it.
practical Β· mixed β Primarily practical: Noam presents original survey research (~6,000 tech workers, 2nd annual run) and moves deliberately toward concrete advice for individuals and leaders. Form is mixed β expository data-walkthrough interleaved with argumentative interpretation (effect-size reasoning, causal claims about managers) and narrative anecdotes/analogies (Terminator, the Cognition/Devon ladder, respondent quotes) used to make statistical findings legible.
Structure
| Part | Role |
|---|---|
| Survey context and methodology | context |
| The AI-identity bifurcation (50% amplified vs. redefined/destabilized/diminished) | core finding |
| Effect-size argument (Cohen's D; AI-identity effect ~3x last year's largest effects) | evidence/methodology |
| Four emotional archetypes (energized, conflicted, disoriented, resentful) | elaboration |
| Burnout and optimism trend data (44.7%β54.7% burnout; 54.8%β48.7% optimism) | core finding |
| Role-recommendation / NPS data (no group is a net promoter of its own role; research/design lowest) | evidence |
| Productivity/quality paradox ('more, not better'; cognitive rot) | core finding |
| Fear ranking (the squeeze and unsustainable pace beat job-loss fear) | evidence |
| Mixed-emotion / 'smiling exhaustion' framing | qualification |
| Functional and company-size breakdowns (design/research worst, data analytics fears job loss most, founders/small companies happiest) | evidence |
| Manager-effectiveness finding | core finding |
| Prescriptive recommendations for employees and leaders | conclusion/practical application |
| Closing note on 'AI guilt' | additional finding |
Key terms
- AI identity shift (amplified / redefined / destabilized / diminished)
Noam's own classification, derived from survey responses, of how a person's sense of professional identity has changed because of AI β a taxonomy with four buckets used as the survey's primary explanatory variable. - smiling exhaustion
Borrowed from a prior guest (Nikhyl Singhal): a state of feeling reborn/energized by building with AI while having 'no off switch,' distinct from classic grim, disengaged burnout. - cognitive rot / 'my brain is rotting'
The erosion of independent judgment and critical engagement that results from habitually accepting AI's first output without applying one's own thinking β a cost of AI use distinct from time-efficiency gains. - effect size / Cohen's D
Used specifically to argue that AI's impact on sentiment is not merely statistically significant (an artifact of large sample size) but practically dominant relative to other variables like role or company size. - the squeeze
The specific, named pressure of rising output expectations ('do more for the same pay') that AI's productivity gains get 'plowed straight back into,' identified as the top-ranked fear in the survey.
Propositions
- explicitdevelopedAI's effect on how people feel about their professional identity is roughly three times as large (by effect size) as the two largest effects found in last year's survey β manager effectiveness and founder happiness.conf: highextends βHow tech workers are feeling in 2026: a workforce splitting in two Β· p1 β Supplies the precise effect sizes (dβ1.55 vs. founder-effect dβ0.56; Ξ²=0.39/0.60) behind the podcast episode's qualitative 'roughly three times as large' claim about the same regression.
- explicitdevelopedSignificant burnout (higher than moderate) rose from 44.7% of respondents in 2025 to 54.7% in 2026; career optimism fell from 54.8% to 48.7% over the same period.conf: highrefines βHow tech workers are feeling in 2026: a workforce splitting in two Β· p2 β Reports 2026 significant burnout at 55.7% versus the podcast retelling's 54.7% for the same year-over-year comparison β a small numeric discrepancy this written report should be treated as the more authoritative figure for.
- explicitdevelopedThe dominant fear among tech workers right now is not losing one's job to AI (ranked second-to-last) but the expectation to do more work for the same pay, followed closely by an unsustainable pace of both work and technology change.conf: highextends βHow tech workers are feeling in 2026: a workforce splitting in two Β· p6 β Adds the near-zero correlation (r=+0.05) between perceived AI job-replacement and layoff worry, sharpening the podcast's qualitative claim about what tech workers actually fear.
- explicitdevelopedAI has increased perceived output quantity ('more, faster') without increasing perceived output quality; the deeper cost is a decline in independent thinking and judgment ('cognitive rot'), reported directly by respondents as 'my brain is rotting, my work feels worse.'conf: highextends βHow tech workers are feeling in 2026: a workforce splitting in two Β· p5 β Adds the 82% productivity topline and additional first-person quotes on judgment/focus loss to the podcast's 'cognitive rot' claim.
- explicitcorroboratedManager effectiveness is the single largest actionable lever on employee wellbeing: highly-effective-manager reports correlate with roughly 65% higher job enjoyment and dramatically lower burnout, yet only ~25% of respondents rate their manager as highly effective and 36% rate their manager as ineffective.conf: highsupports βManagers Are Not Overhead: They Are Infrastructure Β· p4 β Microsoft's finding that manager AI-modeling behavior drives large employee AI-outcome lifts is a specific mechanism supporting the broader survey finding that manager effectiveness is the single largest actionable lever on employee wellbeing/outcomes.extends βHow tech workers are feeling in 2026: a workforce splitting in two Β· p11 β Gives precise manager-effectiveness percentages (25.5% highly effective / 36.5% ineffective) matching the podcast's rounded ~25%/36% figures.
- explicitdevelopedFounders and employees at smaller companies have been the happiest cohort across nearly every measure (optimism, burnout, layoff worry, role recommendation) two years running, yet even founders would not strongly recommend founding a company to others, and nearly half of founders (47%) still report at least moderate burnout.conf: highextends βHow tech workers are feeling in 2026: a workforce splitting in two Β· p10 β Adds founders' exact burnout and recommend-score figures (47% at least moderately burned out, NPS β5) behind the podcast's founder-happiness claim.
- explicitThis survey measures how people feel about their work and about AI, not the objective reality of AI's capability or threat to any given role β the two should not be conflated.conf: high
- implicitOrganizations are simultaneously flattening (fewer managers, more direct reports per manager, 'founder mode') at the same time the data shows manager quality/attention is the biggest lever on wellbeing β a structural tension Noam names but does not resolve.conf: medium
Arguments
AI is having an outlier-level, larger effect on workplace sentiment than any other job characteristic β role, company, company size, or seniority level. strength: medium
Explicit premises:
- Last year's two largest measured effects on job sentiment were manager effectiveness and founder happiness.
- This year's AI-identity-shift classification shows an effect size (Cohen's D) roughly three times as large as those two prior effects.
- The AI-identity bucket correlates strongly and linearly with career optimism, burnout, layoff worry, and role-recommendation scores across the sample.
Implicit premises:
- Effect size (Cohen's D) on a single self-classification question reliably isolates 'AI's effect' as a distinct causal factor rather than a proxy that absorbs other concurrent stressors.
- The single-question AI-identity self-classification is not itself confounded by other salient contemporaneous pressures (layoffs, market conditions) that respondents may attribute to AI simply because AI is the dominant cultural narrative right now.
Inferences:
- If AI-identity is genuinely the dominant single predictor of sentiment, organizational interventions aimed at helping employees process/reframe their relationship to AI should be prioritized over most traditional HR levers.
The single biggest lever leaders have to reduce burnout and improve retention is investing in manager effectiveness. strength: medium
Explicit premises:
- Employees with highly effective managers report ~65% higher job enjoyment and dramatically lower burnout than those with ineffective managers.
- Only ~25% of respondents rate their manager as highly effective, and this figure barely moved from last year.
- The manager-effectiveness effect replicated across both survey years.
Implicit premises:
- Manager effectiveness causes better employee outcomes, rather than merely correlating with them (e.g., via selection β good managers ending up on already-thriving teams, or halo effects where team members in a positive team dynamic rate their manager more favorably).
- Manager effectiveness is a tractable, trainable dimension organizations can meaningfully improve on a timescale that matters, even as the same speaker separately flags a structural trend (org flattening, more direct reports per manager) working against manager capacity and attention.
Inferences:
- Investment in manager training should scale up even as organizations flatten β an implication the argument does not reconcile with the flattening trend it separately raises as a concern.
Narratives
- Noam invokes the Terminator/Skynet movies as an analogy β the very capability that makes AI thrilling to use is the same capability that makes it threatening.
β Illustrates the dual, ambivalent nature of AI excitement and fear underlying the 'clashing emotions' finding. - The Cognition/Devon 'ladder' metaphor (originally from Scott Wu, CEO of Cognition, in an earlier Lenny's Podcast episode): AI capability has climbed from high-school CS student to college intern to junior engineer to senior/staff engineer, and the tech workforce feels the technology 'pulling the rungs out' from under them as it climbs.
β Explains why respondents lower on the seniority ladder feel worse and are less likely to recommend their role to newcomers, and why the erosion is felt as directional/accelerating rather than static. - 'Full gas on neutral' β a car metaphor for individual contributors who are working harder than ever (more PRs, more prototypes, more output) without a corresponding sense of forward progress or impact, amid duplicative, uncoordinated work across the org.
β Explains why ICs, despite increased output, are less likely than VPs/execs to recommend their role β output volume has decoupled from felt impact. - Reference to a prior guest, Jeff from Ramp, who reported that his worst burnout occurred when his velocity was lowest (effort without visible progress).
β Sets up a contrast the episode never fully resolves: the classic 'stagnation causes burnout' theory versus this survey's finding that burnout is rising even as output/velocity rises, complicating a simple velocity-burnout relationship. - Noam's personal admission that he has 'never had as much fun' in his career using these AI tools and has spent excessive hours 'playing about and building,' finding it hard to disengage.
β Grounds the abstract 'addictive technology' claim in first-person testimony, illustrating the enjoyment-yet-exhaustion paradox from the inside. - A respondent quote: 'We're like farmers on the cusp of the industrial revolution and we just don't see a clear path to what's happening.'
β Captures the felt sense of historical discontinuity underlying the 'disoriented' archetype (27% of the sample).
Questions addressed
- β
How is AI reshaping how tech workers feel about their work and identity?
Via the four-way AI-identity taxonomy (amplified/redefined/destabilized/diminished) and the finding that this single variable outweighs role, company, size, and seniority in explaining sentiment. - β
Is AI making tech workers objectively better at their jobs?
Reframed rather than answered affirmatively: people feel they produce more, faster, but not of higher quality, and report a decline in independent judgment ('cognitive rot'). - β What causes burnout in the AI era, and does it fit prior theories (e.g., stagnation-driven burnout)?
Noam offers a partial account (constant learning pressure + 'the squeeze' + no off-switch) but the tension with the velocity-based theory raised via the Ramp anecdote (low velocity β burnout) is named, not reconciled with rising-velocity-yet-rising-burnout data. - β
What is the single biggest actionable lever for improving employee wellbeing?
Manager effectiveness, replicated across two survey years as the largest correlate of enjoyment and (inversely) burnout. - β
Does this survey describe reality (AI's actual capability/threat) or perception (how people feel)?
Noam states explicitly and repeatedly that the survey captures feeling, not objective reality, directly invoking the 'AI confidence theater' critique of overclaiming. - β If sentiment is explicitly disclaimed as not reflecting objective reality, is it still sound to derive concrete behavioral and organizational recommendations from it (go deep not broad; invest in managers; watch the squeeze)?
The recommendations section proceeds directly from the sentiment data to prescriptive action without revisiting the reality-vs-feeling caveat established earlier in the same conversation.
Gaps
- incompleteThe manager-effectiveness finding establishes a strong correlation (via self-report of both manager quality and personal wellbeing) but never disentangles causation from selection or halo effects β e.g., whether effective managers cause better outcomes, or whether already-thriving teams simply rate their managers more favorably, or whether good managers are selectively assigned to already-strong teams.
Read next: Organizational-behavior research isolating causal manager effects (e.g., quasi-experimental studies of manager rotation/turnover on team outcomes), or Gallup's manager-engagement research, for a methodologically stronger version of this claim. - incompleteNoam names a direct structural tension β orgs are flattening (founder mode, more direct reports per manager) at the same time his own data says manager quality/attention is the single biggest lever on wellbeing β but never proposes how to reconcile broader spans of control with the deeper manager investment the recommendations call for.
Read next: Literature on span-of-control limits and manager capacity (optimal team size per manager), or John Cutler's institutional-overload framing, to assess whether 'flatten the org AND invest in managers' is jointly achievable.closed byManagers Are Not Overhead: They Are Infrastructure β Argues directly for one side of the named reconciliation: organizations should weight the emerging evidence that manager behavior drives AI value over the impulse to keep widening spans of control. This only partially closes the gap β it doesn't specify a concrete mechanism for reconciling broader spans with deeper manager investment (see this extraction's own g2 above). - uninformedThe core explanatory variable β the single self-report question classifying 'AI identity shift' as amplified/redefined/destabilized/diminished β is treated as cleanly isolating AI's effect, but a one-question self-classification in a period of concurrent stressors (layoffs, market volatility, general role redefinition) is vulnerable to attribution bias: people may retroactively credit/blame AI for feelings actually driven by other salient factors, simply because AI is the dominant cultural narrative right now. Notably, Noam opens the episode approvingly citing the 'AI Confidence Theater' critique of overclaimed certainty in AI discourse, but doesn't turn that same scrutiny on his own survey's central construct.
Read next: Read the 'AI Confidence Theater' piece (already in this corpus) directly against this survey's underlying methodology/report for confound controls, and look for survey-methodology literature on self-report attribution bias in technology-sentiment research. - incompleteThe 'cognitive rot' / skill-atrophy finding rests entirely on self-perception ('my brain is rotting,' 'my work feels worse') rather than any objective quality measure (e.g., code review defect rates, output audits, before/after task performance).
Read next: Empirical studies attempting to measure AI-assisted output quality or skill atrophy objectively rather than via self-report β e.g., the Simon Willison conversation referenced in the episode, or controlled studies of AI-assisted vs. unassisted task quality.
Related documents
- How tech workers are feeling in 2026: a workforce splitting in two
- Lenny Rachitsky's Post
- KI, utvikling og fΓΈlelser
- Why product feels so hard right now - Charity Ibhadon (WPP)
- AI productivity is burning out your best engineers
- I'm done using AI
- The Hidden Realities of AI Adoption
- How People Are Really Using AI in 2026
- The Psychological Costs of Adopting AI
- SWLW #702: How I created OpenClaw, Your CEO is suffering from AI Psychosis, and more.