AI’s Impact on Software Development Jobs: What the Data Actually Says
Ask software developers whether AI is coming for their jobs and the answers come back fast and confident — either it completely is, or it isn’t at all. The data, though, backs neither answer. What the U.S. Bureau of Labor Statistics (BLS) actually projects is a split outcome: strong growth for software developers, a real decline for programming-only roles, and a job that doesn’t so much disappear as get taken apart, task by task, and reassembled into something different.
This guide sticks to primary sources: the BLS Occupational Outlook Handbook (May 2025 data, 2025–35 projections), BLS’s own published analysis of how it folds AI into those projections, and the 2025 Stack Overflow Developer Survey. Everything below can be traced back to one of those, and if a popular claim couldn’t be verified at its original source, it isn’t in here — when we leave one out, we’ll say so.
The short answer
Short version: AI is eating specific tasks — first-draft code, routine test scaffolding, documentation drafts — not the occupation itself. BLS projects employment of software developers, quality assurance analysts, and testers to grow 10% from 2025 to 2035, much faster than the 3% average for all occupations, adding about 185,400 jobs. The occupation BLS expects to shrink is the narrower one: computer programmers, projected to decline 7% over the same decade, with BLS pointing at AI-driven automation of repetitive programming tasks as a reason why.
Once you see that split, the rest falls into place: the work that’s exposed is the kind that’s mostly typing code, while work that means designing systems, figuring out what was actually asked for, and answering for whether the result is correct is not exposed at all — and AI itself is creating part of the demand for it.
What the BLS numbers actually say
Skip the commentary for a moment and look at the raw figures, all from the BLS Occupational Outlook Handbook:
- 1,905,400 jobs in 2025 — that’s the combined count: software developers (1,717,800) plus QA analysts and testers (187,600).
- +10% projected growth, 2025–35 for the combined occupation — developers +10% (+174,700 jobs), QA analysts and testers +6% (+10,600). All occupations combined: +3%.
- About 106,100 openings a year, on average, across the decade. Some of those are new jobs, but much of it is replacement hiring, as people retire or move on.
- Median pay, May 2025: $135,980 a year for software developers; $104,300 for QA analysts and testers; $134,040 for the combined group.
| Metric | Figure | Why it matters |
|---|---|---|
| Jobs in the occupation (2025) | 1,905,400 | The scale of the field AI is supposedly emptying |
| Projected growth, developers/QA/testers (2025–35) | +10% | BLS’s headline answer: growth, not decline |
| Projected change, computer programmers (2025–35) | −7% | Where the real exposure sits — programming-only work |
| Median pay, software developers (May 2025) | $135,980 | The earning ceiling context (median, not starting pay) |
| Openings per year (average) | 106,100 | New jobs plus replacements — where hiring actually happens |
That median needs careful handling, because it gets abused constantly. A median is the middle of everyone in the occupation: the staff engineer with fifteen years in pulls it up, the new hire pulls it down, and the figure that lands in the middle describes neither of them — and it is neither a starting salary nor a promise. BLS reports that the lowest-paid 10% of software developers earned less than $82,460, and the highest-paid 10% earned more than $214,670 — so a junior developer’s first offer will sit far closer to the bottom of that spread than to the middle. What the median is genuinely good for is showing the size of the ladder. Few occupations offer that much headroom.
Buried in the outlook is a detail most “AI is killing coding” headlines skip. BLS partly attributes the strong developer outlook to AI itself: demand should stay strong, it says, because of “the continued expansion of software development for artificial intelligence (AI), Internet of Things (IoT), robotics, and other automation applications,” plus heavier investment in security software. All of that has to be built, integrated, and maintained by someone.
The split: developers grow, programmers shrink
Software developers and computer programmers sound like the same job, but they are not — and the distance between the two is, more or less, the whole story of AI and coding work.
Software developers design applications and systems: they analyze what users need, plan how the pieces fit together, and own the result. Computer programmers write, modify, and test the code that developers design. BLS projects developers to grow 10% by 2035 — and programmers to decline 7%, sliding from 110,800 jobs in 2025 to a projected 102,700, even though about 4,400 programmer openings a year will still appear as people leave the occupation.
BLS explains the decline in its own words: “Computer programming work continues to be automated,” companies are expected to use technologies “including artificial intelligence (AI), to automate repetitive programming tasks,” and “some higher skilled programming tasks will likely shift to other workers, such as software developers.” The tasks aren’t vanishing from the economy: at the routine end they’re being automated, and at the skilled end they’re absorbed upward into the developer role. That leaves one kind of role exposed — the job that consists of converting someone else’s design into code, in a common language, to a familiar pattern, which is almost exactly the work current AI tools do best.
BLS’s own analysts, coming at it from the other direction, land in the same place. In a 2025 Monthly Labor Review article explaining how AI is incorporated into the projections, they note that although software development is highly exposed to generative AI, the occupation is “unlikely to experience a decline in employment, because robust software needs are expected to support continued demand for its workers.” The reasoning runs in a chain: developers get more productive, software gets cheaper to build, cheaper software gets ordered more often — and AI systems themselves need people to build and maintain them. The same article is candid about uncertainty, to its credit. BLS methods assume technological change proceeds at a pace consistent with past experience, and past displacement, the analysts note, “tends to take longer than technologists typically expect.” So treat the projections as the best official estimate going, not a guarantee. But notice what even the cautious version says: the job changes, the headcount grows.
Task by task: what AI is absorbing, and what it is not
Here’s a rule of thumb worth keeping: occupations don’t get automated. Tasks do. So forget job titles for a second and look at where the line actually falls right now, going by what BLS says developers do and what developers themselves report.
Tasks AI tools now handle or heavily assist:
- First-draft code for well-defined pieces — a function, a component, a routine query, standard boilerplate. BLS’s analysts list “develop, test, and document code” as work GenAI is well suited to augment, and the adoption is hardly hypothetical: 84% of respondents in the 2025 Stack Overflow survey were using or planning to use AI tools in their development process, and 51% of professional developers used them daily.
- Test scaffolding and test-case drafts — generating the likely test cases and edge cases, which a human then reviews and extends.
- Documentation drafts and code explanations. Hand the tool an unfamiliar codebase and it will happily summarize it and draft the doc — a developer still has to correct what it writes.
- First-pass debugging help — paste in an error, get back likely causes and candidate fixes to check.
Tasks that stay human-led:
- Understanding what to build — analyzing users’ needs and turning vague business requirements into a design. BLS lists this first among developer duties, and it’s the part no tool can take from a meeting it did not attend.
- System design — how the pieces fit, what the trade-offs are, what breaks at scale. Someone has to make those calls, and someone has to answer for them.
- Verification and security — deciding whether generated code is actually correct, and actually safe. Developers are holding this line themselves: in the 2025 Stack Overflow survey, more respondents actively distrusted AI output accuracy (46%) than trusted it (33%), and only 3% reported highly trusting it. The top reason developers gave for still asking a human for help was “when I don’t trust AI’s answers” (75%).
- Deployment, monitoring, and planning — the high-responsibility, systemic work, and the survey found developers most resistant to handing it over: 76% did not plan to use AI for deployment and monitoring, and 69% did not plan to for project planning.
Then there’s the frustration data, which explains a lot. The most common complaint in the survey, cited by 66% of developers, is “AI solutions that are almost right, but not quite” — and 45% said debugging AI-generated code is more time-consuming than debugging their own. Code that is almost right is not a finished product; it’s a review task. Somebody with judgment has to do that reviewing, and employers pay for the judgment, not the typing.
Does any of that make the productivity story fake? No — it makes it two-sided. In the same survey, 52% of developers agreed AI tools or agents had a positive effect on their productivity, and among those using AI agents, about 7 in 10 agreed agents reduced the time spent on specific tasks. Both things are true at once. The tools speed up the mechanical parts, and they create verification work that only a competent developer can do. That’s exactly what a job being reshaped, not removed, looks like.
The skills that gain value
When cranking out routine code gets cheap, the skills around the code are what appreciate. Going by the duties and qualities BLS lists for this occupation — and by where the survey shows humans still insisting on control — these are the skills gaining ground fastest:
- System design and architecture. How applications, APIs, databases, and infrastructure fit together, and what breaks when traffic spikes. AI can draft a component all day long, but it doesn’t own the blueprint.
- Verification: testing, debugging, code review. The occupation literally includes QA analysts and testers, projected to grow 6%, and every developer now reviews machine output as well as human output. Being the person who catches the “almost right” bug before production is a promotable skill.
- Security. BLS expects rising investment in security software to drive developer demand. Generated code with a subtle vulnerability is worse than no code, and employers know it.
- Requirements and communication. BLS lists analytical and communication skills among the occupation’s important qualities: explaining a technical problem to a nontechnical customer, and asking the questions that reveal what the software actually needs to do.
- Working with AI tools well. Prompting precisely, fencing the output in, and checking it like a skeptic. This is a real skill now — and per the survey’s trust numbers, the checking matters more than the prompting itself.
- Domain knowledge. A developer who understands healthcare, finance, or logistics brings something a general-purpose tool does not: context about why the software exists.
If you are presenting these skills to an employer, our guide to the skills to put on your resume shows how to evidence them instead of just listing them.
If you are a junior developer or a student
The honest version first: the pressure sits on the bottom rung of the ladder. The simple, well-specified coding tasks that used to train beginners are exactly the tasks AI now drafts in seconds, and the programmer occupation’s projected 7% decline is the statistical shadow of that shift. That doesn’t make the career a bad bet — a 10% growth projection and roughly 106,100 openings a year say otherwise — but the way in has changed, concretely:
1. Learn the fundamentals anyway — they are now your quality control. You can’t review code you couldn’t have written. Data structures, how the web actually moves data, how databases store and retrieve it: this is the knowledge that lets you catch the “almost right” output instead of shipping it.
2. Build complete projects, not tutorial fragments. A small application you designed, built, tested, and deployed — and can explain decision by decision in an interview — proves the skills AI cannot fake for you: design, debugging, follow-through.
3. Practice reviewing AI output deliberately. Generate code, then interrogate it: What does it do with bad input? Where is it insecure? What breaks at ten times the data? Bring that critique habit to interviews; it is the job now.
4. Learn testing early. Write tests for your own projects. Verification is a growth area inside the occupation, and most juniors neglect it — an easy way to stand apart.
5. Get fluent with Git, collaboration, and reading other people’s code. Teams hire for this. It rarely shows up in “learn to code” marketing, but it shows up in nearly every real job.
6. Aim your applications at developer work, not programming-only work. Postings that ask you to understand requirements, design features, and own outcomes are the growing category. Pure code-conversion work is the shrinking one. For the wider picture of how beginners break into remote roles, see our guide to entry level remote jobs.
One more route worth knowing: many developers working today arrived from other fields entirely. If you are weighing a switch into tech from a non-technical career, our breakdown of the best jobs for career changers covers how that path usually works and what it pays.
What this means if you are hiring or managing developers
The same data cuts both ways for employers: the ones getting value from AI tools are the ones treating output as a draft that a developer verifies, not as finished work that skips review. The survey numbers — 46% distrust, 66% burned by “almost right” code — describe your team’s experience too if review standards slip. And the BLS projections describe your hiring market: a growing occupation with about 106,100 openings a year means the developers with design, security, and verification skills will have options. Replacing them with a tool subscription is not one of the scenarios the data supports.
Frequently asked questions
The official projections say no. BLS expects software developer employment to grow 10% from 2025 to 2035 — much faster than the 3% average — partly because AI applications themselves create developer demand. What AI is replacing is specific tasks: routine, well-specified code production. The occupation built around those tasks alone, computer programmers, is projected to decline 7%.
Yes. BLS projects software developer employment to grow 10% from 2025 to 2035, and the May 2025 median pay was $135,980. About 106,100 openings a year, on average, is a strong profile alongside that. The caveat: the median covers all experience levels — the lowest 10% of developers earned under $82,460 — and the entry point now rewards design, testing, and verification skills earlier than it used to.
The repetitive, well-scoped ones: boilerplate and first-draft code, routine test scaffolding, documentation drafts, and first-pass debugging suggestions. BLS says the same about the programming occupation — companies are expected to use AI “to automate repetitive programming tasks.” Tasks involving requirements, system design, security decisions, deployment, and accountability for correctness remain human-led, and that’s also exactly where developers report resisting AI most strongly.
Because the occupations are defined differently. Developers analyze needs and design systems; programmers mainly write and test code to others’ designs. AI is strongest at the second job and weakest at the first, and BLS expects some higher-skilled programming tasks to shift to developers. The work moves up the skill ladder rather than disappearing.
BLS lists a bachelor’s degree in computer and information technology or a related field as the typical entry-level education for this occupation, and notes some employers prefer a master’s for certain positions. That is the typical path, not a legal requirement — but unlike the no-experience occupations covered elsewhere on this site, this one is degree-typical, and beginners should plan with that reality in mind.
Start from the spread, not the headline figure. The May 2025 median was $135,980 across all experience levels; the lowest-paid 10% earned less than $82,460, and a first job typically lands in the lower part of the range before rising with experience. QA analysts and testers — a common entry point into the same field — had a $104,300 median, with the lowest 10% under $61,440.
Conclusion
The honest summary of AI’s impact on software development jobs is narrower than the headlines running in both directions. AI is genuinely absorbing the routine layer of the work — BLS projects a 7% decline for programming-only roles and says so in plain language. It is also, by BLS’s own account, expanding the market for developers: +10% growth, roughly 106,100 openings a year, and demand partly driven by the need to build AI systems themselves. The developers exposed to the change are the ones whose job was only ever the typing. The ones who design, verify, secure, and explain are the ones the next decade is short of.
Two famous claims we left out
Two widely repeated claims about AI and coding appear nowhere above, and that is deliberate. The first is the statement, attributed to Google CEO Sundar Pichai on an Alphabet earnings call in late 2024 and repeated everywhere since, that AI now generates more than a quarter — over 25% — of Google’s new code; we did not confirm that figure on a published source from Google itself in this session, so we do not use it as evidence here. The second is the result of GitHub’s 2022 controlled experiment on Copilot, which reported that developers using the tool completed a task 55% faster; we did not re-open the original study to check it today, so we do not cite it either. Both claims may or may not hold up — until a claim is checked at its source, it stays out.
How we check our numbers. Every pay and outlook figure in this guide was read on the BLS Occupational Outlook Handbook pages listed below on October 23, 2026. Company and survey claims are used only from their own official pages, and all figures are re-checked when the article is updated.
Sources
– Stack Overflow Developer Survey 2025, AI section
Pay and outlook figures are from the sources above, checked October 2026. Projections are estimates, not guarantees — BLS updates them annually, and company tools and practices change faster than statistics do.
About the author: Tajammul Abbasi researches and compiles U.S. job-market data from the Bureau of Labor Statistics and O*NET for HirePath USA, focusing on entry-level and remote roles. Every national pay figure in this guide comes from the U.S. Bureau of Labor Statistics or O*NET. Last reviewed: October 23, 2026.
