We do not call any tech job title safe from AI, and we will not list safe ones. Federal projections show growth for software developers and a decline for computer programmers from 2025 to 2035 [1][2]. The Bureau's stated reasons are about tasks, such as repetitive programming [2]. So check your tasks, not your title.
This page walks through what the labor data says and what a developer trial and its later update found. Then it shows how to read your own week as an engineer.
What does the labor data say about tech jobs?
The Bureau of Labor Statistics publishes ten-year projections for each occupation. Two of them sit side by side in tech, and they point in opposite directions.
For software developers, quality assurance analysts and testers, it projects 10 percent growth from 2025 to 2035 [1]. It calls that much faster than the average for all occupations. It also projects about 106,100 openings a year over the decade [1].
For computer programmers, it projects a 7 percent decline over the same years [2]. That is 8,100 fewer jobs, from a base of 110,800 [2]. It still projects about 4,400 openings a year [2].
Those are projections, not promises. They cover a whole occupation across the whole country. They say nothing about your team, your employer or your role.
The Bureau is open about the limits too. In a 2025 article on AI and its projections, it wrote that "projecting future employment involves substantial uncertainty" [3]. It added that the timing and scale of many potential impacts of generative AI are too uncertain to be reflected in its projections [3].
Why do two close tech titles get opposite projections?
The Bureau gives its reasons in a note on each occupation. Both notes are about the work being done [1][2].
Its note on programmers opens this way: "Computer programming work continues to be automated, helping computer programmers to become more efficient in some of their tasks" [2]. It expects many companies to use technologies, including AI, to automate repetitive programming tasks [2]. It adds that some higher skilled programming tasks will likely shift to other workers, such as software developers [2].
Its note on developers reads differently. It ties demand to the continued expansion of software development for AI, the Internet of Things, robotics and other automation applications [1]. It also expects organizations to invest more in software that protects their networks [1]. That could raise demand for developers who build security software, and for testers who write and run software tests [1].
Put the two notes next to each other. In our reading, the word that carries the weight is "repetitive." The Bureau expects companies to automate repetitive programming tasks [2]. It names software built for AI and robotics as a source of demand, and says security needs could add to it [1].
The 2025 article says the same thing in general terms. AI is expected to mainly affect occupations "whose core tasks can be most easily replicated by GenAI in its current form" [3].
Here is the catch, and this part is our advice, not the Bureau's. Do not assume your title tells you which note fits your week. Ask whether your days go to repeated tickets or to design calls. Then follow the tasks.
Do AI coding tools make experienced engineers faster?
The measured answer has shifted, and the researchers who measured it urge caution [4][6]. Start with their 2025 trial.
Researchers ran a randomized controlled trial with 16 experienced open-source developers [4]. The developers completed 246 tasks in mature projects they knew well. They averaged five years of prior experience on those projects [5]. Each task was randomly assigned to allow or disallow AI tools [4].
Before starting, the developers forecast that AI would cut their completion time by 24% [4]. With AI allowed, tasks took 19% longer [4]. Afterward, the developers still estimated that AI had made them 20% faster [4].
Read that result with care, as the authors do. They call it a snapshot of early-2025 tools in one setting [4]. They do not claim their developers or projects represent most software work [4]. They say AI tools may well be useful in other contexts, such as for less experienced developers or in an unfamiliar codebase [4].
The same researchers have since added to it. In February 2026 they reported on a second study, begun in August 2025 with newer tools [6]. Its raw results lean the other way. For ten developers returning from the first trial, the estimate was 18% less time with AI [6]. The range ran from 38% less time to 9% more [6]. For newly recruited developers, the estimate was 4% less time, with a range from 15% less to 9% more [6]. Both ranges include no change at all.
The authors do not lean on those numbers. They call the new data an "unreliable signal" [6]. More developers declined to take part because they did not want to work without AI [6]. The authors say that likely pushes their estimate of the speedup down [6]. They believe developers are likely more sped up in early 2026 than in early 2025 [6]. They also call their data "only very weak evidence" for the size of that change [6].
Two lessons still hold for your own work:
- Your sense of speed is not a measurement. In the first trial, developers believed they were faster when they measured slower [4]. The update says self-reported speedups can be quite unreliable [6].
- One result does not transfer everywhere, or last. The first trial covered people working in code they knew deeply, with early-2025 tools [4]. The authors now think developers are likely faster than that trial showed [6].
Neither lesson says AI tools are bad at code. They say you should check what a tool did on your work, with your numbers.
Which tech tasks should you check in your own week?
All of them, one at a time. A title-level answer will not help you. A task-level one can.
List what you did last sprint. Be specific. "Coding" is not a task. "Add a field to the invoice API and migrate the table" is.
Then rate each task on three things:
- AI reach. Could a coding agent do this now, with the repo access it would really have?
- Stakes. What breaks if it is wrong? Who gets paged, and who pays?
- Human edge. Does the task lean on context the agent lacks? Think of past incidents, unwritten conventions or what the customer meant.
Give each task one call. Hand it off, pair with a tool, own it, or cut it. Our page on how the task audit works lays out the steps.
Pay extra attention to review. If an agent opens pull requests on your team, someone decides whether they merge. Do not treat a green check as proof that the change is right. Ask what the tests cover and what they leave out. Finding what the tests do not cover is a skill you can practice.
That is the center of The Hard-to-Replace Engineer, our kit for mid-level software engineers. You review a simulated agent's pull request and judge twelve findings in its report. Then you write the review and a merge decision. An answer key grades every finding. The case is invented and nothing runs on your machine.
The kit does not promise a job, a raise or a promotion. No kit can. It gives you a clear map of your tasks and a finished review you can show.
Want to try the method first? The free practice case has a case brief and an AI first pass to check. It comes from another role's kit, but the habit of checking an AI's claims is the same one.
What else do people ask about tech jobs and AI?
Is any tech job title safe from AI?
We do not call any title safe. Federal projections show growth for software developers and a decline for computer programmers over 2025 to 2035 [1][2]. The Bureau's stated reasons are about tasks [1][2]. It says projecting employment involves substantial uncertainty [3].
Do federal job projections account for AI?
Only in part. The Bureau of Labor Statistics says the timing and scale of many potential impacts of generative AI are too uncertain to be reflected in its projections [3]. Treat every projection as a national estimate, not a promise.
Why does the data treat programmers and developers differently?
They are separate occupations in the federal data. The Bureau projects a 7 percent decline for computer programmers from 2025 to 2035. It says programming work continues to be automated [2]. It projects 10 percent growth for developers and testers [1].
Do AI coding tools always save time?
Not in every measurement. In an early-2025 trial, experienced developers took 19% longer with AI tools on projects they knew well [4]. In February 2026, the same researchers said developers are likely sped up more than their early-2025 estimate showed. They called their newer data only very weak evidence [6].
Sources
- "Software Developers, Quality Assurance Analysts, and Testers," Occupational Outlook Handbook, U.S. Bureau of Labor Statistics, 2026. https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm
- "Computer Programmers," Occupational Outlook Handbook, U.S. Bureau of Labor Statistics, 2026. https://www.bls.gov/ooh/computer-and-information-technology/computer-programmers.htm
- "Incorporating AI impacts in BLS employment projections: occupational case studies," Christine Machovec, Michael J. Rieley and Emily Rolen, Monthly Labor Review, U.S. Bureau of Labor Statistics, 2025. https://www.bls.gov/opub/mlr/2025/article/incorporating-ai-impacts-in-bls-employment-projections.htm
- "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity," METR, 2025. https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
- "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity," Joel Becker, Nate Rush, Elizabeth Barnes and David Rein, arXiv (Cornell University), 2025. https://arxiv.org/abs/2507.09089
- "We are Changing our Developer Productivity Experiment Design," Joel Becker, Nate Rush, Tom Cunningham, David Rein and Khalid Mahamud, METR, 2026. https://metr.org/blog/2026-02-24-uplift-update/