What Economists Expect AI to Do to Tech Hiring
Forecasts about AI and jobs tend to swing between panic and hype. A useful corrective is to look at what professional economists actually said when asked, in a structured survey, with numbers attached.
Indeed Hiring Lab published one in September. Its Q3 2026 Labor Market Outlook Survey, run with Pulsenomics, was fielded September 8 to 16 and drew 123 panelists. These are forecasts, not measurements, and they come from a panel of economists rather than hiring managers. They are still a more disciplined signal than a viral post.
The headline: modest, not apocalyptic
Asked about AI’s net effect on employment, 45% of the panel said “minor negative,” 25% said “minor positive,” and 21% expected no net effect. On productivity over the next three years, 47% expect a “moderate” boost and 27% a “substantial” one. Only 3% expect a “transformative” gain, and 2% expect none.
If you read that as “AI changes everything next year,” the panel disagrees. If you read it as “nothing will change,” it also disagrees.
Software development appears on both lists
The most useful detail for tech candidates is in the occupation breakdown. The survey asked where AI-driven employment losses and gains are expected to be largest.
- Largest expected losses: Administrative Assistance, Software Development, and Data & Analytics.
- Largest expected gains: IT Infrastructure Operations & Support, Software Development, and Data & Analytics.
Software Development and Data & Analytics are on both lists. The honest reading is not “software engineers are doomed” or “safe.” It is that the field is expected to reshuffle internally: some work shrinks, some grows, and the survey does not say which tasks fall on which side. The panel also named Software Development among the fields with the largest expected declines in job postings.
The current baseline
The survey sits on top of a market that is already soft for software. In its October 2026 snapshot, Indeed’s postings index for software development is about 79 on a scale where February 2020 equals 100, the lowest of the five sectors in its chart. The same snapshot puts AI-related postings at 7.3% of all postings, a share measured by keywords, as Indeed’s methodology note explains.
Two things are true at once: overall software demand is below its pre-pandemic level, and the share of postings that mention AI keeps climbing. Candidates who move toward the second trend have a different market from those who do not.
Indeed’s snapshot also notes pockets of strength. Advertised wages for IT systems and solutions roles grew 4.7% year over year, and data center postings are up 147% from June 2024 levels. Demand in tech has not vanished; it has moved toward infrastructure.
The wage warning is about degrees, not skills
The survey also asked about wages. Its diffusion index for AI-driven displacement risk is 55.6 for college-educated workers, where 50 is neutral, and 48.1 for workers without a college degree. The index measuring whether AI hurts college or non-college wages more fell from 47.3 to 42.4, which Indeed describes as the most pronounced quarter-over-quarter change in the survey.
Be careful with this one. It is an expectation of pressure on degree-holding workers as a group. It says nothing about any particular specialization, and tech roles that require a degree are not automatically the ones at risk.
What to do with it
- Stop reading forecasts as verdicts. The panel’s own split, 45% minor negative against 25% minor positive, shows how much disagreement remains.
- Look at tasks inside your role. If the same occupation can appear on both the gain and loss lists, the useful question is which of your tasks grow with AI: evaluation, deployment, data quality, infrastructure.
- Check what a posting actually asks for. A keyword count includes plenty of roles that only mention AI. Roles where AI is the daily work are a smaller, more specific group, and that is the group we curate on the AIfirst.jobs feed.
The next outlook survey will show whether the panel’s views are hardening. Until then, treat these numbers as informed guesses with a stated margin of disagreement.