Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteNo one has directly measured the share of AI-written code across all software outside GitHub, and no current source claims to. The closest broad answer comes from JetBrains’ Developer Ecosystem Survey 2026. Professional developers reported that, on average, about 47% of the work code they produced in the previous month was fully generated by AI agents, and about 38% was written by the developer with some AI assistance. That is a self-reported estimate for one population’s output. It is not a share of any codebase, and the two figures should not be added together.
Why there is no single number outside GitHub
GitHub-based studies can infer authorship from public repository artifacts. Almost everything outside GitHub, including private repositories, company monorepos, other hosting platforms and internal tools, has no equivalent public trail. To estimate AI’s share there, someone has to ask the people who write the code, or ask the organization that owns it. Each approach counts something different, so the numbers that come back are not interchangeable.
Four choices change the answer every time: who is counted (all professional developers, startup teams, one company’s engineers, or the contributors to a set of public projects), what unit is measured (recent work output, committed code, or the share of an existing codebase), how “AI involvement” is defined (fully generated by an agent, or any assistance at all), and what period is covered (the last month, a survey fielding window, or six years of commit history).
The sources and what each one counts
The table below lists the main contemporary figures with the population and unit behind each. Read the last two columns before comparing any two rows.
| Source and date | Population | What is counted | Reported figure | Evidence type |
|---|---|---|---|---|
| JetBrains Developer Ecosystem Survey 2026 (fielded May–July 2026) | More than 15,000 professional developers worldwide, reweighted to the global developer population by region, employment status, programming language and familiarity with JetBrains products | Share of code produced for work in the previous month, reported in bands | Averages of about 47% fully agent-generated, about 38% AI-assisted and about 27% fully manual | Self-reported survey |
| Supabase, State of Startups 2026 | Surveyed startups | Share of the startup’s existing codebase written by AI | 61% say more than half the codebase is AI-generated; 40% place it at 76–100%; 2% report zero | Startup self-report survey |
| Sonar, State of Code Developer Survey 2026 (summary dated January 8, 2026) | Developers surveyed by Sonar | Share of code they commit, with generated and assisted code combined | 42% AI-generated or AI-assisted | Developer self-report |
| Science study (published 2025) | More than 30 million GitHub commits from 160,097 developers in six countries, 2019–2024 | Python functions in GitHub projects classified as AI-written | 29% of Python functions in the United States estimated AI-written | Classifier inference from repository artifacts (based on the published abstract) |
| Anthropic, internal report (May 2026) | Anthropic’s own codebase | Code merged into Anthropic’s codebase | More than 80% authored by Claude as of May 2026 | Company-reported internal figure |
| GitHub enterprise survey with Wakefield (fielded February 26–March 18, 2024) | 2,000 non-student, non-manager employees at companies with at least 1,000 staff; 500 each in the U.S., Brazil, Germany and India | Use of and attitudes toward AI coding tools at work | More than 97% had used AI coding tools at work at some point; share of code generated: not stated | Survey measuring adoption, not code share |
JetBrains: the broadest measure of developers’ own output
JetBrains asked respondents what share of the code they produced for work last month was “fully generated by AI agents,” “written by you with some AI assistance,” or “fully written by you without any AI assistance.” Answers came in bands (0%, 1–20%, 21–40%, and so on up to 81–99%, 100%, and “I don’t know”), so the published averages depend on bucket midpoints.
That method has a direct consequence. The three averages add up to about 112% (47 + 38 + 27), which is why JetBrains warns that averages within a group can exceed 100%, and that self-reports may not always be accurate. Use each category as its own estimate. Do not combine them into a headline like “85% AI-written.” The figure describes what developers believe they produced, and it is the only one in this list that spans a broad mix of employers and roles rather than a single company or a single kind of team.
Supabase: a startup codebase measure
Supabase’s startup survey asks a different question. It measures the share of a startup’s codebase that is AI-written, not the work output of one month. The numbers are striking, but they describe startup respondents who chose to answer the survey. The page reviewed did not provide enough methodological detail to treat them as representative of all startups, let alone all software teams.
Sonar: committed code, with assistance included
Sonar’s developers estimated that 42% of the code they commit is AI-generated or AI-assisted. Two features set this apart from JetBrains. The unit is committed code, and the definition merges generated and assisted code into one category. Its companion finding is also useful: 38% of respondents said reviewing AI-generated code took more effort than reviewing code written by human colleagues.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #3
The Science study: what repository analysis can see
The Science study used a classifier on more than 30 million GitHub commits from 160,097 developers in six countries, covering 2019 to 2024. It estimated that AI wrote 29% of Python functions in the United States. This is the most rigorous kind of measurement in the list, but it observes only what is in the studied GitHub projects and languages. It cannot see private repositories, other platforms, or software written in other languages. Its value here is in showing how far repository-based inference can go, and where it stops.
Anthropic: a company-specific figure
Anthropic reported in May 2026 that Claude authored more than 80% of the code merged into Anthropic’s own codebase. The company also said its typical engineer was merging eight times as much code per day in Q2 2026 as in 2024. Anthropic itself cautions that lines of code measure quantity rather than quality, and that the volume figure overstates true productivity gains. Treat the 80% figure as an example from one company with a large internal tooling investment. It is not an industry estimate.
GitHub’s 2024 survey: adoption, not share
The 2024 enterprise survey fielded by Wakefield for GitHub is often cited near code-share numbers, but it measured something else. More than 97% of respondents said they had used AI coding tools at work at some point. That tells you how widely tools are used. It does not tell you what proportion of code they produced.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why the figures conflict less than they seem
The Supabase startup share, the Sonar committed-code share, the Anthropic merged-code share and the Science Python-function share are not competing measurements of the same thing. Each answers a different question. A startup’s existing codebase, a developer’s recent work, a company’s merged code, and a sample of public GitHub functions are four different units. The sensible comparison is to ask which unit matches your question, not which number is largest.
Recommended Free Tools
Best Value
Measuring it yourself outside GitHub
If you need a figure for your own organization, the same choices apply. Define the unit before you ask anyone anything.
- Pick one unit. Choose recent work output, committed changes, or the share of the existing codebase. Do not mix them in one chart.
- Write the categories out. Separate “generated by an agent with little or no edit,” “written by a person with AI assistance,” and “written without AI.” Keep these distinct, as JetBrains did.
- Use bands and report the method. Bands reduce false precision. Publish the midpoints you used and do not sum categories into a total.
- Sample deliberately. Report who answered, their roles and company size, and how many declined to say.
- Cross-check with a disclosure signal. If your team tags AI-assisted changes in commit messages or pull request templates, count those tagged changes as a floor. This measures disclosure, not authorship, so label it that way.
What the number does and doesn’t tell you
A high AI share says that a lot of code was produced with AI involvement. It does not say the code is better, worse, cheaper, or that fewer people are needed. Several sources in this list point to review burden and quality questions, and Anthropic’s own caution about lines of code applies to every volume measure. Use these figures to understand where code comes from and how it is reviewed. Do not use them as evidence of productivity, quality or headcount.
Bottom line for readers
The most defensible current answer is a range of reported experiences rather than a single percentage. In JetBrains’ 2026 survey, professional developers reported that about 47% of their recent work code was fully agent-generated and about 38% AI-assisted. Startup codebases, committed code and one company’s merged code can show higher shares, but those figures describe narrower populations and different units. Outside GitHub, no one has measured the total share directly, and any claim that they have should be checked against the unit and population it actually covers.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
