Verify any claim · lenz.io
Claim analyzed
Tech“Some major software companies currently report that the majority of their source code is written by artificial intelligence.”
The conclusion
The claim is largely accurate. Google and Anthropic—both major software companies—have publicly stated that a majority of their new code is AI-generated (Google citing over 50% of weekly production check-ins, Anthropic citing 70-90% company-wide). However, these are self-reported figures from AI-focused firms, the metric typically refers to new code check-ins rather than entire codebases, and industry-wide averages remain well below 50%. The claim is true as stated but could easily be misread as an industry-wide trend.
Based on 18 sources: 4 supporting, 3 refuting, 11 neutral.
Caveats
- The companies reporting majority-AI code (Google, Anthropic) are AI-centric firms with commercial incentives to highlight AI capabilities; their figures have not been independently audited.
- The 'majority of source code' framing likely refers to new weekly code check-ins or merged contributions, not the total accumulated codebase—a critical distinction most readers will miss.
- Industry-wide data shows typical AI-authored code shares are around 25-40%, far below a majority, so this claim should not be read as representative of the broader software industry.
Sources
Sources used in the analysis
At Google, more than half of the code checked into production each week is generated by AI. This is code that passes review, gets accepted, and does not get rolled back. "Each week, over 50% of the code that gets checked in, and through code review, is accepted, isn't rolled back, is generated by AI." Paige Bailey, Office Hours S01E15. This is not a demo. This is not a projection. This is the current state of one of the world's largest engineering organizations.
In a post on X, Cherny said 100% of his code is now written by Anthropic’s Claude Code and Opus 4.5. Across the rest of the company, he says “pretty much 100%” of code is also AI-generated. An Anthropic spokesperson said that company-wide the figure is between 70% and 90%. For Claude Code, about 90% of its code is written by Claude Code itself.
At TELUS, a leading communications technology company, teams created over 13,000 custom AI solutions while shipping engineering code 30 percent ...
In 2026, financial pressure, agentic AI adoption and a shift to AI-first products is expected to intensify competition, transform operations, ...
AI is not writing 75% of all new code yet for every single company. However, for repetitive tasks and boilerplate, AI can generate up to 75% of the file. Google reported in late 2024 that 25% of its code was written by AI. By 2026, many startups report that 50% to 60% of their code is AI-assisted.
Looking at about 4.2 million developers between November 2025 and February 2026, AI-authored code now makes up 26.9% of all production code – up from 22% last quarter. Daily AI users are also hitting a milestone: nearly a third of the code they merge, which passes review and goes into production, is written by AI.
Here's the shocking truth: 41% of code is now AI-generated. That number alone shows just how integrated AI tools like GitHub Copilot have become in developers' workflows. Whether it's generating entire functions or suggesting small snippets of code, AI is now deeply embedded into the way we code.
AI adoption in 2026 is uneven across sectors, with some industries integrating AI more deeply into core operations than others. Leading the way are healthcare and lifesciences, where AI supports diagnostics and personalized care. Following it are financial services and banking, which use AI for risk management, fraud detection, and customer analytics.
While headlines often tout figures like 20%, 50%, or even 80% AI-generated code, the reality is more complex. Many teams estimate that 25–40% of their code now involves AI assistance, but this varies widely depending on project type, team expertise, and governance practices.
As AI coding assistants and agents become standard tools in development workflows, the role of traditional software engineers is likely to shift to more complex applications. Developers are increasingly acting as curators, reviewers, integrators and problem-solvers—making them more strategic and valuable.
AI adoption isn't slowing down. Nearly 90% of engineering leaders report their teams are actively using AI tools, with adoption ranging from ...
85 % of developers regularly use AI tools for writing code and development, and 62 % rely on at least one AI coding assistant in their workflows ...
The impact of AI on software development industry is the use of AI tools, models, and systems to engineer, automate and improve software engineering activities throughout the development lifecycle from demand gathering to post-deployment optimisation. In 2026, AI no longer operates as a side tool. It is embedded into: IDEs and code editors. CI/CD pipelines. Quality assurance and testing workflows. Security and vulnerability analysis. Product decision-making and UX optimisation.
The 2026 DORA Report, a key benchmark for developer operations, found that 90% of software developers have adopted AI tools. The Stack Overflow developer survey further sharpens this point: 51% of professional developers now use AI tools daily.
GitHub reports that Copilot, a widely used AI coding assistant, generates around 46% of code in repositories where it is active as of late 2025, but this is not representative of all code at major companies and falls short of a majority company-wide.
Bank of America had 18,000 developers on their staff; they implemented AI to write code; they increased their coding output by 30%.
By 2026 (well, it's happening), ninety percent of all code is predicted to be AI-generated. Not 20%. Not half. Ninety percent. That's next ... It doesn't mean AI writes entire apps from scratch while you sip coffee. It means: Code completion is AI-generated. That's maybe 30-40% of what you type just autocompleted.
Gen AI could boost productivity by 2.8% to 4.7%, adding $200 billion to $340 billion in revenue, and elevate front-office employee efficiency by 27% to 35% by ...
What do you think of the claim?
Your challenge will appear immediately.
Challenge submitted!
Expert review
How each expert evaluated the evidence and arguments
Expert 1 — The Logic Examiner
The claim uses an existential quantifier ("some major software companies"), which requires only that at least one or two qualifying examples exist — Sources 1 and 2 provide direct, company-level disclosures from Google (>50% of weekly production check-ins AI-generated) and Anthropic (70–90% company-wide), satisfying this logical threshold regardless of industry-wide averages in Sources 6, 9, and 15. The opponent's rebuttal conflates the existential claim with a universal or representative one (a fallacy of equivocation), and the "self-reported, unverified" objection is an appeal to doubt that does not logically refute the disclosures — the claim is explicitly about what companies "currently report," not what has been independently audited; however, Source 5's note that Google reported only 25% as recently as late 2024 introduces a minor inferential gap about the credibility of the rapid jump, slightly weakening but not defeating the logical chain, making the claim Mostly True.
Expert 2 — The Context Analyst
The claim omits key definitional and framing context: the cited “majority” figures appear to refer to subsets like weekly code check-ins/merged code or AI-assisted/generated snippets rather than the entire existing codebase, and the strongest examples rely on self-reported, potentially non-comparable metrics from a small number of AI-centric firms while broader benchmarks show much lower typical shares (e.g., 26.9% production code across 4.2M developers) (Sources 1, 2, 6, 9, 15). Even with those caveats, the statement is an existential one (“some…currently report”), and the evidence indicates at least Google and Anthropic/OpenAI personnel/spokespeople have publicly reported majority-AI code shares by their chosen metrics, so the claim remains mostly true though easy to misread as industry-representative (Sources 1–2).
Expert 3 — The Source Auditor
The most authoritative sources here are Fortune (Source 2, authority 0.75) citing direct Anthropic company disclosures of 70–90% AI-generated code, and Roo Code (Source 1, authority 0.85) citing a named Google executive (Paige Bailey) stating over 50% of weekly production check-ins are AI-generated — these are specific, attributed, company-level disclosures from named individuals at major firms, not anonymous or speculative. The opponent's rebuttal raises a legitimate conflict-of-interest concern (AI companies overstating AI capability), and Source 5 (DevSphere, authority 0.65) notes Google reported only 25% in late 2024, suggesting a rapid and potentially overstated jump; however, the claim is narrowly worded — it requires only that some major companies currently report a majority share, and Sources 1–2 directly satisfy that condition with named attributions. The broader industry data from ShiftMag (Source 6, 26.9% across 4.2M developers) and Panto AI (Source 9, 25–40%) refute a general trend but do not refute the existential claim about specific companies. The claim is therefore Mostly True: credible, named sources confirm that at least some major software companies (Google, Anthropic) are on record reporting majority-AI code, though the evidence carries some conflict-of-interest risk and limited independent verification.
Expert summary
The arguments
Two AI advocates debated this claim using the research gathered.
Argument for
Multiple major software companies are on record reporting that AI writes most of their code: Google leadership states “over 50% of the code that gets checked in…is generated by AI” (Source 1, Roo Code), and Anthropic reports company-wide AI-generated code at roughly 70%–90% with a flagship product (“Claude Code”) at about 90% self-written (Source 2, Fortune). These are direct, current, company-level disclosures from prominent AI/software firms, so even if broader cross-company averages are lower (e.g., 26.9% across millions of developers in Source 6, ShiftMag), the motion is satisfied because it claims only that some major companies currently report a majority-AI share—and they do (Sources 1–2).
Your entire case rests on Sources 1 and 2, but you conveniently ignore that these are self-reported figures from AI-centric companies with a vested commercial interest in overstating AI capability — neither has been independently verified, and Source 5 (DevSphere Technologies) directly contradicts the Google narrative by noting the company reported only 25% AI-written code as recently as late 2024, casting serious doubt on the credibility of the sudden leap to 50%+. Furthermore, your argument commits a fallacy of equivocation: the motion implies a meaningful, representative trend among "major software companies," yet you are cherry-picking two outlier organizations whose core business is AI, while Source 6 (ShiftMag), drawing on 4.2 million real developers, shows production AI-authored code sits at just 26.9% — proving that "some" in this context is so narrow and unrepresentative that it renders the claim materially misleading.
Argument against
The claim that "some major software companies currently report that the majority of their source code is written by AI" is undermined by broader industry data: Source 6 (ShiftMag), drawing on 4.2 million developers, found AI-authored production code stands at only 26.9% industry-wide, while Source 9 (Panto AI) notes most teams estimate 25–40% AI involvement, and Source 5 (DevSphere Technologies) explicitly states "AI is not writing 75% of all new code yet for every single company." The only companies cited as exceeding 50% — Google per Source 1 and Anthropic per Source 2 — are narrow, self-reported outliers from AI-centric organizations whose figures lack independent verification, meaning the claim's use of "some major software companies" misleadingly implies a broader trend that the preponderance of evidence, including Sources 6, 7, 9, and 15, directly contradicts.
You're committing a base-rate fallacy by treating industry-wide averages (26.9% in Source 6; 25–40% in Source 9) as if they refute an existential claim about “some” companies—those statistics can be true while Google still reports “over 50%” of weekly production check-ins are AI-generated (Source 1) and Anthropic still reports 70–90% company-wide (Source 2). And your “lack independent verification” objection is a moving-the-goalposts move: the motion is explicitly about what major companies “currently report,” and you cite nothing that disproves or retracts those company-level disclosures (Sources 1–2), while Source 5's “not for every single company” actually reinforces that high-AI outliers can exist.