Will AI replace software engineers? What the data says in 2026

No, AI is not going to replace software engineers as a profession, but it is already replacing a large share of the coding they used to do by hand. The U.S. Bureau of Labor Statistics (BLS) rates software developers as very highly exposed to AI, yet it still projects their jobs to grow 10% by 2035, while jobs for computer programmers, who mainly write code to someone else’s design, are projected to shrink 7%.

That split is the real answer. AI is good at producing code. It is much weaker at deciding what should be built, fitting it into a messy existing system, judging whether it is safe and correct, and being accountable when it breaks at 2 a.m. Below are the official numbers, what AI can and can’t do in software, which tech roles are most and least at risk, why new graduates feel the change first, and how to stay ahead of it.

The numbers: very exposed, still growing

In August 2026 the BLS published official AI exposure ratings for 831 occupations, alongside its job projections for 2025 to 2035. The ratings combine five research measures, including real-world AI usage data from Anthropic and Microsoft, and sort each occupation into one of four roughly equal groups: low, moderate, high or very high exposure.

Software is one of the most exposed fields in the data. Every computer occupation the BLS lists is rated very high, from software developers and web developers to database administrators and help-desk staff. Yet their projected futures differ sharply:

Pay is the median for May 2025. All seven occupations are rated very high for AI exposure.

Software developers are a large group: about 1.72 million jobs in 2025, projected to reach about 1.89 million in 2035, with roughly 95,300 openings a year. The BLS gives the reasons in its outlook: continued growth in software for AI, the Internet of Things, robotics and other automation, plus more spending on security software. In other words, AI is one of the reasons demand for developers is projected to grow, not only a threat to it.

Computer programmers tell the opposite story. Their jobs are projected to fall from 110,800 to 102,700. The BLS says programming work keeps being automated, that companies are expected to use AI to automate repetitive programming tasks, and that some higher-skilled programming work will likely shift to software developers. All of the roughly 4,400 programmer openings expected each year come from people retiring or leaving, not from new demand.

Pay reflects where the value sits. The median software developer earned $135,980 in May 2025, more than two and a half times the $50,980 median for all occupations. The median programmer earned $100,390.

Will AI replace programmers? Coding is the most exposed task

The research that measures real AI use, rather than guessing at it, puts coding at the top of the list.

Anthropic measured which work tasks its Claude models are actually used for, in professional settings, and compared that with what each job involves. Computer programmers came out on top of every occupation, with 75% of their tasks covered, ahead of customer service representatives and data entry keyers. A separate Anthropic study of 500,000 coding conversations found that 79% of conversations with its coding agent, Claude Code, were automation, where AI carried out the task directly, against 49% on its general chat app. The most common work was web-facing: JavaScript and TypeScript made up 31% of queries and HTML and CSS 28%, with building user-interface components the single largest task.

Microsoft analyzed 200,000 anonymized conversations with its Copilot assistant and scored 785 occupations on how much of their work AI was used for, and how well. Web developers rank 21st, quality assurance analysts and testers 53rd, computer programmers 81st and software developers 120th. Information security analysts rank lower, at 205th. The pattern is consistent: the closer a job is to turning a clear specification into code, the higher it ranks.

That is why the BLS projects decline for programmers but growth for developers. When AI drafts the code, the job of a person who only writes code to order shrinks. The job of a person who decides what the code should do, and checks that it does it, grows.

What AI can and can’t do in software engineering

Developers have adopted AI faster than almost any other profession. In Stack Overflow’s 2025 Developer Survey, 84% of respondents used or planned to use AI tools in their work, and 51% of professional developers used them daily. Yet trust is low: 46% said they distrust the accuracy of AI output, against 33% who trust it. The most common frustration, cited by 66%, was AI answers that are almost right but not quite.

Measured productivity is less clear-cut than the hype. In a 2025 randomized trial by the research group METR, 16 experienced open-source developers worked on 246 real issues in projects they knew well. They expected AI to speed them up by 24%. With AI allowed, they actually took 19% longer, and afterwards still believed AI had made them about 20% faster. METR stresses that this was a small study of one setting with early-2025 tools and now treats the result as historical, since tools have improved. It still shows something lasting: on complex, familiar codebases, the time AI saves on typing can be lost on prompting, waiting and fixing.

Here is how a software engineer’s work splits:

• Most exposed: writing routine code. Boilerplate, CRUD endpoints, user-interface components, data transformations, unit tests, simple scripts and documentation. AI produces a fast first draft; a person must check it.
• Partly exposed: debugging, refactoring and code review. AI can explain an error, suggest a fix and flag problems in a pull request, but finding the real cause in a large system and deciding what is safe to change still needs someone who understands it.
• Least exposed: deciding what to build and owning the result. Turning vague business needs into requirements, designing systems and trade-offs, security and reliability decisions, handling incidents, and agreeing priorities with people. These rest on context, judgment and accountability.

The work is moving up that list. More of an engineer’s day goes to specifying, reviewing and integrating, and less to typing.

Which tech jobs are most and least at risk?

Because every computer occupation is rated very high, the useful question is not whether a tech job is exposed but whether demand for it is growing anyway.

The roles under the most pressure are the ones built around a single, well-defined output: computer programmers, computer user support specialists (−3.5%), and network and computer systems administrators (−4.1%). Front-end and web work is also heavily exposed, which matches both Anthropic’s and Microsoft’s data, though web developer jobs are still projected to grow 3.8%.

The roles in the strongest position combine exposure with fast-rising demand. Data scientists (+34.6%) and information security analysts (+21.0%) are growing many times faster than the 3% average for all jobs, and software developers themselves sit well above average. Our own scoring, which uses the same method as our free test on a typical week, puts a software developer at 70 out of 100, in the very high band but below jobs that are almost entirely routine information work:

A real developer’s score depends on how their week splits. Someone who spends most of the day writing code to spec will score higher; a lead who spends half the week on design, reviews and incidents will score lower. These three careers in our index sit closer to the decisions AI can’t own:

For jobs outside tech that hold up best, see the 25 most AI-proof jobs.

Why junior developers feel it first

If AI is not cutting software jobs overall, why does the market feel so hard for new graduates? Because the first effect of AI shows up in hiring, not firing, and mainly at the entry level.

Stanford economists Erik Brynjolfsson, Bharat Chandar and Ruyu Chen tracked payroll data from ADP covering millions of U.S. workers. In their August 2026 update, employment of 22–25-year-olds in the two most AI-exposed groups of occupations fell about 11% between November 2022 and June 2026, while the same age group in the least exposed jobs grew about 10%. They put young workers in exposed jobs 19% below where they would be had they kept pace, and name software development among the occupations affected. Experienced workers show no comparable gap, and the change works mainly through fewer hires rather than more layoffs. The authors call these early, descriptive signals rather than proof that AI is the cause, and note that the pattern holds even when tech firms and computer jobs are excluded.

Anthropic found a similar but smaller signal in government survey data: no rise in unemployment for highly exposed workers since late 2022, but an estimated 14% drop in the rate at which young workers find jobs in exposed occupations, a result it describes as just barely statistically significant. There was no such drop for workers over 25.

The logic is simple. The tasks companies used to give junior engineers, such as small features, bug fixes, tests and documentation, are exactly the tasks AI now drafts. Senior engineers, whose value lies in judgment and knowledge of the system, become more productive with AI rather than less needed. For the wider picture of which jobs AI is changing fastest, see What jobs will AI replace?

Will AI replace programmers in 10 years?

The official ten-year outlook says the profession will grow, but its shape will change. BLS projections run from 2025 to 2035 and already factor in automation. They show programmer jobs shrinking and developer, security and data jobs growing. Searches like “will AI replace programmers in 20 years” or “by 2050” go beyond what anyone can credibly forecast; what the evidence shows is which tasks go first.

In software, that order is clear. Routine code goes first, then much of testing and first-pass debugging. Design, integration, security and accountability go last, because someone has to decide what the software should do and answer for what it does. As AI writes more code, the amount of software in the world, and the amount that needs to be designed, secured and maintained, keeps growing. That is the BLS’s own reasoning for projecting developer growth.

What will change within ten years is how engineers start their careers. Juniors will be expected to arrive able to read, test and fix AI-generated code, not only write their own, and to show judgment earlier than before.

How to AI-proof a career in software

The rule for engineers is the same as for doctors and lawyers: let AI handle production, and own the decisions and the result. Here is how each type of task holds up:

For software careers, five moves matter most:

1. Use AI daily, and review everything it writes. The engineers who gain most treat AI as a fast junior colleague: useful, but never trusted without tests and review. Learn where your tools fail, especially on security and edge cases.
2. Move toward design and ownership. System design, architecture, reliability and handling incidents are the least exposed parts of the job. Volunteer for them early instead of waiting for seniority.
3. Go deep in a domain. An engineer who understands payments, healthcare records or industrial systems is much harder to replace than one who only knows a framework. Domain knowledge is what turns vague requirements into the right software.
4. Build skills in growing specialties. Security, data, cloud infrastructure and AI engineering combine high exposure with fast-growing demand. Our index rates cybersecurity analysts at 40 and cloud infrastructure engineers at 48, well below the 70 we give a typical developer week.
5. If you are early in your career, build proof. With fewer junior openings, real projects, open-source contributions and evidence that you can ship and maintain working software count for more than a degree alone.

For a full 90-day plan to build the skills that hold their value, see our guide to the skills AI can’t replace. The same pattern of AI taking the drafting while people keep the responsibility plays out in law, as we show in Will AI replace lawyers?