If your workforce plan sorts jobs into AI exposed and AI proof, the newest federal projections say to stop. The Bureau of Labor Statistics projects 5.9 million new jobs from 2025 to 2035, with total employment rising from 170.3 million to 176.2 million. That is growth of 3.5 percent, well below the 10.9 percent recorded from 2015 to 2025. Alongside the forecast, BLS launched new AI exposure categories covering 831 occupations. The headline is what did not appear: across those occupations, SHRM reporting on the release notes that BLS found no relationship between an occupation's AI exposure and whether its employment is projected to grow or shrink. For CIOs, CISOs and VP IT Operations leaders carrying firmware, test, network and security requisitions, that finding changes the hiring question. The question is no longer which roles AI will take. It is which tasks inside each role now require senior judgment, and how you prove a candidate has it before you make the offer.
What did the BLS projections actually find about AI and job growth?
Start with pace. A slower decade is the baseline for every plan. Healthcare and social assistance are projected to add more than 2.2 million jobs and account for about 37 percent of all new jobs through 2035. Utilities are projected to grow fastest by rate at 9.8 percent, but from a small base, adding 58,800 jobs, with nearly all of the growth in electric power generation, transmission and distribution as AI power demand rises. Professional, scientific and technical services are projected to grow 8.6 percent and add 926,700 jobs. Computing infrastructure providers, data processing, web hosting and related services are projected to grow 25.1 percent and add 120,400 jobs. Computer and mathematical occupations are projected to grow 7.3 percent. Office and administrative support moves the other way, down 4.0 percent and down 752,100 jobs. None of that sorts cleanly by exposure. BLS is explicit in its exposure methodology that a category is not a forecast of employment growth or decline, not a wage forecast and not a worker replacement estimate. The categories combine three theoretical measures of how well AI matches occupational abilities and tasks with two observed measures drawn from real AI usage mapped to tasks and work activities. Treat them as a map of where tasks may change, not as a prediction of which teams shrink.
Why does the pay premium look like a seniority premium?
Posting data show where the pressure actually concentrates. Indeed Hiring Lab, analyzing millions of United States postings that list an annual salary, found advertised pay in the most AI exposed occupations up about 46 percent since 2021, against 25 percent in the least exposed. The gap tracked closely for the first year after ChatGPT launched and widened notably from 2024. Once the changing mix of occupations is held constant, the post ChatGPT premium is 5.7 percent. Comparing each job title with its own past, it is 4.7 percent. Holding the seniority mix constant inside each occupation, it narrows to 2.4 percent and is not statistically significant at the 5 percent level. At senior level, more exposed pay grew 45 percent from 2021 to mid 2026 against 28 percent for less exposed pay, a gap of about 17 points. At mid level the gap is about 12 points. At entry level it is about 2 points. Inside the most exposed occupations, the entry share of salaried postings fell from 29 percent to 10 percent between 2021 and 2026, while the senior share rose from 22 to 47 percent. Salaried postings skew senior, so read the levels with care. The direction still holds. Employers are posting fewer junior seats in exposed work and competing harder for people who can direct tools, check outputs and own the decision.
What does a three year high in tech postings change for IT operations hiring?
Demand is already pressing on the same senior pool. CompTIA, analyzing Bureau of Labor Statistics data and employer postings, reported 272,040 new openings for IT occupations in September and 625,633 active postings, the highest total in more than three years. New postings were up 16 percent from September 2025 and active postings up 22.4 percent. Software developers and engineers drew 47,598 new openings, up 8.3 percent from August, followed by systems engineers, tech support specialists, data analysts and DevOps engineers. Twenty three of the 30 highest volume technology occupations gained month over month. Active postings calling for AI related skills totaled 349,821, up 32,327 from August and up about 282 percent since May 2024. Dedicated AI and machine learning job title postings reached 13,636 active postings, nearly triple the May 2024 level. Technology sector employment fell by about 10,350 positions in September even as postings rose, with tech unemployment at 3.3 percent against a national rate of 4.2 percent. Postings up and employment flat is not a contradiction. It is a queue. Approved requisitions for cybersecurity, network and infrastructure, SRE and NOC, test and validation, firmware and embedded, field service, ERP support and data center roles are competing for experienced people who are employed, selective and able to wait. In that queue, a posting that describes tasks poorly or screens only for keywords will sit while a better specified search closes.
How should CIOs and talent leaders hire when exposure labels mislead?
Plan by tasks, then hire by evidence. Take each open technical role and separate the work into three buckets: routine execution that tooling can assist, oversight work that needs a person to validate outputs, and stakeholder judgment that only a senior operator can provide, such as incident calls, vendor tradeoffs and compliance sign off. Score candidates against the second and third buckets in writing. A firmware engineer who has brought up hardware, debugged across the hardware and software boundary and documented the failure modes is a different hire from a longer keyword list. A test and validation specialist who can show defect evidence, environment coverage and release judgment is scorable. A cybersecurity analyst who can evidence triage, containment and clear escalation notes is scorable. That scoring only works if the search reaches beyond inbound. The person you need is rarely applying. That is the work UPPER is built for: sourcing channels worked systematically across the United States and Canada, every candidate scored against the same written evidence, and a ranked shortlist your own recruiters operate and your hiring manager can judge quickly.
What should you check before the next requisition opens?
Run five checks on Monday. Does the requisition name the tasks that need judgment? Is the seniority honest? Did the hiring manager agree to a written evidence rubric before sourcing starts? Does the search cover sourcing channels beyond the posting itself? What is the date of the next decision, and who owns it? The projections give you no shortcut from an exposure label to a hiring plan. The postings say senior judgment is where pay and competition are moving now. Teams that write for judgment, search widely for proof of it and decide fast will fill the seats that keep systems running.