Nobody on your team has been laid off. But the team of nine is now six, the two people who left were not replaced, and every line of next year’s roadmap has an agent in it. You are a senior engineer with a good rating and a full backlog. The question in front of you is not whether you will lose your job this quarter. It is what to do in the period while the choice is still yours.

Most commentary on AI and work offers one of two answers: panic, or reassurance. The labor-market data support neither. They show a specific pattern of risk, concentrated in a specific place, and they say something useful about which moves have paid off for people in contracting occupations before. Software engineering is the clearest current example, and the lessons apply well beyond it.

What the data show, and what they do not

Start with what has not happened. Stanford researchers tracking payroll data through June 2026 report no widespread, economy-wide job displacement. What they do find is narrower: employment of workers aged 22 to 25 in highly AI-exposed occupations is about 19% below where it would be had it kept pace with their peers in less-exposed work, a gap that was 15% a year earlier. This is a descriptive measure, not an estimate of what AI caused, and it comes mainly from fewer hires, not more separations. Where AI is used to complement workers, employment is flat or rising, particularly for the experienced [1].

Other trackers are more cautious still. Yale’s Budget Lab, using national household survey data through August 2026, finds no clear evidence of labor-market disruption associated with AI [2]. New York Fed economists concluded in May 2026 that job postings give little indication of a distinct AI-driven decline in demand [3]. In the same bank’s regional survey, 4% of service firms said they had laid off workers because of AI in the previous six months, up from 1% a year before. About 15% said they had hired fewer people than they otherwise would have, and about 13% said they had hired more [4].

Attribution is contested. A Stanford policy brief lists interest-rate rises and pandemic over-hiring as rival explanations for weak hiring in exposed fields, while judging that AI had become a more plausible direct cause by 2024 [5]. The authors of the payroll study themselves report that, with their broadest controls, the decline becomes statistically significant only from 2024 [6].

One finding recurs across sources and matters most for an individual. The adjustment is running through the hiring door. In Canada, central bank staff estimate that the gap in job-finding rates between workers in fully AI-exposed occupations and unexposed ones widened from 2.2 percentage points before the pandemic to 13.9 in 2025, while the gap in separation rates barely moved [7]. People in seat have mostly stayed in seat. The difficulty begins for anyone who has to get back in.

Software, the worked example

Software engineering is not disappearing. Indeed reports that US software development postings grew almost 15% between February 2025 and mid-2026 while overall postings fell 7%, though the category still sat about 27.5% below its level of February 2020. The composition of that recovery is the telling part: of the net increase in postings between May 2025 and May 2026, 71% came from senior roles and 37% from jobs with AI in the title [8]. The Bureau of Labor Statistics still projects employment of software developers to grow 15.8% between 2024 and 2034, against 3.1% for all occupations [9].

The pressure is at the entrance and in the queue. New-graduate hiring at twelve large technology companies is down roughly 65% from 2019, and the ratio of experienced candidates to open roles in technology is the highest in a decade [10]. Employment in computer systems design has fallen 5% since late 2022 while total US employment rose about 2.5%. Over the same period, a Dallas Fed analysis finds, wages have been rising in AI-exposed occupations that place a high value on tacit knowledge and experience [11].

Put together, the occupation is being re-sorted, not removed. There are fewer seats at the bottom, more candidates for each opening, and a premium on the judgment that comes with experience. An experienced engineer in a stable seat is well placed. The same engineer on the open market is competing in the most crowded field in ten years.

The cost is in the re-entry

The government’s survey of displaced workers puts numbers on that. Some 3.3 million long-tenured US workers lost jobs between 2023 and 2025, about 746,000 more than in the previous period, and 44.4% of them because their position or shift was abolished. In January 2026, 66.1% were working again, almost exactly the rate two years earlier. Finding work was not the problem. Finding equivalent work was: among those who had lost full-time jobs and were back in full-time work, about 49% were earning as much as before, down from about 62% in the previous survey. Age mattered too. Re-employment was 72.9% for those aged 25 to 54 and 57.3% for those aged 55 to 64 [12].

People who change jobs still see faster pay growth than people who stay, 4.8% against 3.0% in ADP’s September 2026 data [13]. That payroll series does not separate people who chose to move from people who were pushed. Set beside the displacement survey, though, it supports a reasonable inference: the penalty attaches to involuntary exit, not to moving. A move made from a job is a negotiation. A move made after losing one is a search under pressure, and half the people in that position take a pay cut.

Direction matters more than speed

What separates those who recover from those who do not? A Goldman Sachs analysis of four decades of US data, as reported in the financial press, found that workers displaced from technology-disrupted occupations took about a month longer to find a new job and lost more than 3% of real earnings on re-employment, where workers displaced from more stable occupations lost little [14]. The month is not the story. The analysts attribute most of the loss to occupational downgrading: displaced workers were more likely to land in more routine jobs that used fewer analytical and interpersonal skills. Losses were smaller for those who moved into roles with higher analytical content, and for those who retrained [15].

That research covers earlier waves of technology, and the smaller losses were observed mainly among younger, college-educated, urban workers, so it should be read as a guide to mechanism and not a forecast. But other evidence points the same way. A New York Fed study of 1.6 million federally funded training spells found that AI-exposed workers earn high returns from retraining in general, and that those who retrain specifically toward AI-intensive occupations earn 29% lower returns than those who pursue broader training [16]. Chasing the technology that displaced you is not the highest-yield move.

Nor are the new AI roles a plan for most people. US job postings asking for AI skills reached 80,597 in August 2026, with median advertised pay of $147,500 [17]. Listings for forward-deployed engineers, one of the most discussed new roles, rose from roughly 300 to more than 5,000, and the firm tracking them describes the role as demanding senior engineering skill [18]. These jobs are real and well paid. They are also few, and they are not entry points.

The encouraging finding is about capacity. Brookings estimates that around 70% of the US workers most exposed to AI, 26.5 million of 37.1 million, are in jobs with a high capacity to manage a transition. The 6.1 million with high exposure and low capacity are mostly in clerical and administrative roles [19]. If you are an experienced professional in an exposed field, the evidence says you are among those best equipped to move. The question is where to.

What to do while the choice is still yours

  • Read your own role first. List what you do in a week. Mark which parts a tool now does acceptably and which parts people still bring to you: the ambiguous problem, the trade-off, the customer, the decision someone has to own. The second list is your direction.
  • Map adjacent roles by content, not by title. For an engineer, that may mean architecture, reliability, security, platform work, technical product management, customer-facing engineering, or leading a team. Ask of each: does it use more judgment and more contact with people than what I do now, or less?
  • Try the internal door before the external one. A sideways move inside your employer, made before a role is cut, keeps your track record and your pay. Internal candidates are hired at the highest rate of any source.
  • Start doing the next job before you hold it. Take a piece of the adjacent work now, and record what it produced. Evidence of the direction is what makes the move credible.
  • Retrain broadly. The returns were higher for general capability than for a narrow bet on AI occupations.
  • Keep the cost of re-entry low. A current record of outcomes, relationships that are not dormant, and two or three people who have seen your recent work and would say so. Structured search support works when it builds both skill and motivation: in a review of 47 evaluated programs, the odds of finding work were 2.67 times higher, but only for programs that combined the two [20].

Two earlier articles go further into the mechanics: how to choose an adjacent move and which of your assets are gaining and losing value.

Persona Map is built for this kind of planning: a private workspace to record what your work has produced, map the people who have seen it, and chart adjacent paths, with an AI advisor that works from that record. The broader point stands without it. The data do not say experienced people in exposed roles are about to be swept out. They say that the expensive moment is an unplanned exit, and that the direction of the next move decides most of what follows. Both are things you can act on while you still hold the seat.

Sources

  1. Stanford Digital Economy Lab. “Canaries in the Coal Mine?” August 2026 update. Link
  2. The Budget Lab at Yale. “Tracking the Impact of AI on the Labor Market.” September 15, 2026 update. Link
  3. Federal Reserve Bank of New York, Liberty Street Economics. “Do Job Postings Show Early Labor Market Effects of AI?” May 2026. Link
  4. Federal Reserve Bank of New York, Liberty Street Economics. “Businesses Are Using AI to Transform Work, Not Cut Jobs.” September 2026. Link
  5. Stanford Institute for Economic Policy Research. Policy brief on AI and jobs by Mahoney, McEntarfer & Wahal. July 2026. Link
  6. Stanford Digital Economy Lab. Follow-up analysis on interest rates and timing. February 2026. Link
  7. Bank of Canada. Staff analysis of AI exposure and labor-market flows. August 2026. Link
  8. Indeed Hiring Lab. “AI and Job Postings: From Destruction to Creation.” July 8, 2026. Link
  9. US Bureau of Labor Statistics. The Economics Daily: artificial intelligence, information technology and employment, 2024–34. Link
  10. SignalFire. State of Talent Report 2026. Link
  11. Federal Reserve Bank of Dallas. J. Scott Davis. “AI Is Simultaneously Aiding and Replacing Workers, Wage Data Suggest.” February 24, 2026. Link
  12. US Bureau of Labor Statistics. Worker Displacement: 2023–2025. News release, August 2026. Link
  13. ADP. National Employment Report, September 2026. September 30, 2026. Link
  14. Yahoo Finance. Report on a Goldman Sachs research note on technology displacement. April 2026. Link
  15. Fortune. Report on the same Goldman Sachs research note. April 6, 2026. Link
  16. Federal Reserve Bank of New York. Staff Report 1165, on returns to retraining for AI-exposed workers. 2025. Link
  17. CFO.com. Report on Lightcast job-posting data for AI skills. 2026. Link
  18. LinkUp. Job-listing analysis of AI-related roles. 2026. Link
  19. Brookings Institution. “Measuring US Workers’ Capacity to Adapt to AI-Driven Job Displacement.” 2026. Link
  20. Songqi Liu, Jason Huang & Mo Wang. “Effectiveness of Job Search Interventions: A Meta-Analytic Review.” Psychological Bulletin 140, 2014. Link