Robots Are Better Than People home
Subscribe

Will AI Replace Programmers?

AI tools now write much of the routine code that programmers used to type by hand, and federal projections show the narrow programmer job title shrinking. The work that remains is reviewing, securing and taking responsibility for that code.

Our readShrinking title, shifting workOur read: AI is absorbing routine coding and the programmer title is projected to decline, but someone still has to check, secure and answer for the code.

A ticket lands in the queue: a date field breaks when users in another time zone submit a form. You pull the branch, read the stack trace, find the faulty conversion, write the fix, add a test and open a pull request. Until recently, nearly every keystroke in that loop was yours. Now an AI assistant can propose the fix before you finish reading the trace. So the real question is which parts of this loop still need you.

What the job numbers say about programmers

The U.S. Bureau of Labor Statistics tracks computer programmers as its own occupation, separate from software developers. Its employment projections for computer programmers show 110,800 jobs in 2025 and 102,700 projected for 2035. That is a -7.3% change, while all occupations are projected to grow 3.5%.

The field still averages 4,400 openings a year, and the median annual wage was $100,390 in 2025. A bachelor's degree is the typical entry point.

The BLS Occupational Outlook Handbook page for computer programmers lists 121,200 jobs in 2024 and an outlook of -6% (Decline) for 2024 to 2034. Keiser University reports that the handbook, updated August 27, 2026, now projects a 7% decline. Second Talent notes the handbook ties that fall to companies using AI to automate repetitive programming tasks.

Yahoo News reported that programmer employment hit its lowest level since 1980 and that the drop lines up with the arrival of OpenAI's ChatGPT. Lining up is not the same as proving cause.

Programming tasks: what AI handles and what stays with you

TaskWhat AI does todayWhat still needs a person
Writing routine functions and boilerplateGenerates new code at scale at Google and MicrosoftDeciding whether the code fits the system and the business rule
Fixing bugsModifying software to correct errors is a leading use of Claude, per Anthropic's Economic IndexConfirming the root cause, not just the symptom
Security reviewProduces working code that often carries flaws, per Veracode researchCatching vulnerabilities before release
Running changes against live systemsAgents can execute commands directlyGuarding production data, as the SaaStr Replit incident showed
Signing off on contributionsCan draft patchesLegal and technical accountability, per Linux kernel rules
Multi-step coding workClaude Code automated more of its conversations than Claude.ai, per AnthropicSetting goals and judging when the agent is wrong

Where AI is already writing production code

Google CEO Sundar Pichai said in October that more than 25% of new code was written by AI, CNBC reported. On a later earnings call he put the figure at well over 30%, according to Entrepreneur. At Google Cloud Next in April 2026, he said AI writes 75% of Google's new code, with engineers checking it before it ships, per Elite Brains.

Microsoft CEO Satya Nadella said up to 30% of the company's code was written by AI, TechCrunch reported. TechCrunch also cautioned that it is unclear how either company measures AI-generated code.

Layoffs now cite AI directly. GitLab cut around 14% of its workforce in June 2026 as it leans on AI agents, Programs.com lists. Across the U.S., AI was cited in 87,714 announced job cuts through May 2026, 22% of all layoffs, Uvik reports. Those are employer-stated reasons for all jobs, not programmer counts.

Anthropic's labor market study ranks Computer Programmers at the top of its list, with 75% coverage.

Programming skills gaining value as AI spreads

  • Secure code review. Veracode tested 80 curated coding tasks across more than 100 large language models and found Java's security failure rate over 70 percent. A Spring 2026 update covering more than 150 models found newer models had not closed the gap.
  • Production safety. After its agent deleted a user's database, Replit's CEO announced dev/prod database separation and a planning-only mode. People who design those guardrails matter.
  • Accountable contribution. The Linux kernel adopted a formal AI policy in April 2026 requiring an Assisted-by: tag, Review Commit reports. The kernel's guidelines say to "expect additional scrutiny in proportion to how much of it was generated."
  • Licensing literacy. The Ninth Circuit ruled on September 16, 2026 in Doe v. GitHub, Haynes Boone explains, but Lexology notes it did not settle whether reproducing copyrighted code is lawful.

What to do next, depending on who you are

  • Student: The narrow programmer title is shrinking. Aim your degree at engineering work that includes design, security and review, not only typing code.
  • Mid-career programmer: Become the person who reviews AI output, writes the tests and owns production safety. That is where accountability sits.
  • Employer or client: Treat AI code as a draft. Veracode's findings and the Replit incident show what happens when nobody checks.

Myths vs reality for programmers

MythReality
AI always makes coders fasterA METR trial of 16 seasoned developers found they took 19% longer with AI tools, IT Brew reported. METR later wrote: "Speedups now seem likely, but changes in developer behavior make our new results unreliable."
AI has wiped out engineering hiringSignalFire data shows engineers making up a larger share of new hires, TechCrunch reported, and Nvidia's engineers are busier than ever since adopting agentic AI.
Code that runs is safe codeVeracode found AI produces functional code that still introduces security flaws.
Software jobs are finishedCNN ran an April 2026 analysis titled "The demise of software engineering jobs has been greatly exaggerated," HeroHunt notes.

Our verdict: the programmer title shrinks, the responsibility grows

We think AI will not erase the people who make software work, but it is already eating the routine coding that defined the programmer title. The BLS projection points down, and big employers say AI writes a growing share of their code. Our read is that programmers who move toward review, security and ownership of production systems keep their footing. Those who only translate specs into syntax face the steepest pressure.

The bottom line: AI writes more of the code every year, but someone still has to check it, secure it and sign their name to it. Make that someone you.

Questions people ask

Will AI replace computer programmers?

BLS projects computer programmer jobs to fall from 110,800 in 2025 to 102,700 in 2035, a -7.3% change. AI is taking routine coding work, but someone still has to review, secure and take responsibility for the code.

How much code is AI writing at big tech companies?

Google CEO Sundar Pichai said in April 2026 that AI writes 75% of Google's new code, with engineers checking it before it goes live. Microsoft CEO Satya Nadella said up to 30% of his company's code was written by AI, though TechCrunch noted it is unclear how these figures are measured.

Does AI make programmers faster?

A METR trial with 16 seasoned developers found they took 19% longer when using AI tools. METR later said speedups now seem likely, but that changes in developer behavior made its newer results unreliable.

Is AI-generated code safe to ship?

Veracode tested more than 100 large language models and found AI produces functional code that still introduces security flaws, with Java failing over 70 percent of the time. A Spring 2026 update covering more than 150 models found newer models had not closed the gap.

Are programmers still being hired?

Yes. BLS expects 4,400 openings a year on average, and SignalFire data reported by TechCrunch shows engineers making up a larger share of new hires.

Sources

Every number, quote and example on this page was checked by script against the sources below before publishing: quotes must match the source word for word, and figures must appear in a source or in BLS data. Our verdict is editorial judgment. Spotted an error? Tell us.

  1. U.S. Bureau of Labor Statistics: Employment Projections, Computer programmers (15-1251)
  2. Is AI Replacing Computer Science? (keiseruniversity.edu)
  3. AI in Software Development Statistics for 2026 - Second Talent (secondtalent.com)
  4. Employment for computer programmers in the U.S. has plummeted to its lowest level since 1980—years before the internet existed (yahoo.com)
  5. Computer Programmers (bls.gov)
  6. Satya Nadella says as much as 30% of Microsoft code is written by AI (cnbc.com)
  7. AI Is Taking Over Coding at Microsoft, Google, and Meta (entrepreneur.com)
  8. AI-Generated Code Stats 2026: How Much Is Written by AI? | Elite Brains (elitebrains.com)
  9. Microsoft CEO says up to 30% of the company's code was written by AI | TechCrunch (techcrunch.com)
  10. Economic Index report: Economic primitives (anthropic.com)
  11. Labor market impacts of AI: A new measure (anthropic.com)
  12. Economic Index: AI's impact on software development (anthropic.com)
  13. AI coding tools might actually be slowing you down (itbrew.com)
  14. METR (@METR_Evals) on X (x.com)
  15. AI was supposed to kill engineering jobs, but new data suggests they're the most resilient | TechCrunch (techcrunch.com)
  16. Tech Layoffs and AI: The 2026 Reality Check (herohunt.ai)
  17. AI Job Displacement Statistics 2026: Layoffs | Uvik Software (uvik.net)
  18. List of Companies Announcing AI-Driven Layoffs - Programs.com (programs.com)
  19. What Veracode's CWE Breakdown Reveals About AI-Generated Code | PlayerZero (playerzero.ai)
  20. AI-Generated Code Poses Major Security Risks in Nearly Half of All Development Tasks, Veracode Research Reveals (businesswire.com)
  21. Ninth Circuit Rejects DMCA Copilot Theory | News | Haynes Boone (haynesboone.com)
  22. Ninth Circuit Limits DMCA Claims Against Generative AI: Key Legal Findings and Implications for IP Teams - Lexology (lexology.com)
  23. An AI Contribution Policy That Says What It Means (reviewcommit.substack.com)
  24. Kernel Guidelines for Tool-Generated Content — The Linux Kernel documentation (cdn.kernel.org)
  25. Linux kernel now allows AI-generated code, as long as you take "full responsibility" for any bugs (pcguide.com)
  26. Replit agent database deletion | KPATH (kpath.ai)
  27. Replit’s New Release Addressed Most of The Challenges We Hit Vibe Coding. But Is ‘Prosumer’ Vibe Coding Really Ready for Commercial Apps Yet? (saastr.com)

The White-Hat Brief

Weekly: what the machines learned, and how to stay ahead

One email a week. Verified stories, plain-language defenses, no hype. Unsubscribe anytime.