AI Coding Stats 2026: 50 Data Points From 7 Surveys

AI coding tool adoption reached 84–91% across four major surveys in 2025–2026, yet trust in AI accuracy dropped to 29% among Stack Overflow respondents — down 11 percentage points from 40% the prior year. This guide aggregates 50 verified statistics from 11 primary sources, flags where surveys contradict each other and why, and provides original analysis of the velocity-versus-risk tradeoff that now defines the AI coding era.

The headline adoption numbers are not wrong — they are just measuring different things. Stack Overflow's 84% counts developers who "use or plan to use" AI tools. DORA's 90% measures professionals who have "adopted AI at work." DX's 91% comes from a dataset of 85,350 developers already inside organizations tracked by DX's engineering intelligence platform — a self-selected sample skewed toward AI-forward teams. Understanding those definitional gaps is the prerequisite for any honest benchmarking conversation about where your team sits.

This guide covers adoption rates (10 stats), tool preferences and market share (10 stats), a proprietary source-divergence table where surveys directly contradict each other, productivity impact (8 stats), code quality and technical debt signals (8 stats), security and compliance data (8 stats), enterprise versus indie patterns (6 stats), and regional and demographic splits (7 stats). Every statistic is traced to a primary-source URL. For teams evaluating tooling decisions, our AI coding IDE landscape guide and the companion per-tool cost calculator provide the operational complement to these statistics.

Four major surveys converge on near-universal adoption but diverge by as much as 7 percentage points depending on how they define "using AI for coding." None of these numbers are wrong — they are measuring genuinely different things, and conflating them produces misleading benchmarks.

The broadest count is Stack Overflow's: 84% of all developers use or plan to use AI tools in their development process, up from 76% in 2024, per the Stack Overflow 2025 Developer Survey (fielded May 29–June 23, 2025; published August 2025 with an extended analysis published December 29, 2025). The 2026 Stack Overflow survey had not been published as of May 2026 — these are the latest available figures. Of that 84%, 51% of professional developers use AI tools daily, and 47.1% of all respondents (including learners and students) use them daily.

JetBrains sits one point higher on the general adoption headline: 85% of developers regularly use AI tools for coding or development, per the JetBrains State of Developer Ecosystem 2025 (October 2025). Of that group, 62% rely on at least one dedicated AI coding assistant, agent, or AI-native code editor — versus general AI chat tools. JetBrains' January 2026 follow-up survey (published April 2026) found 90% of developers regularly use at least one AI tool at work for coding, with 74% having adopted a specialized AI tool — a dedicated coding assistant rather than a general-purpose LLM chat interface.

DORA's framing is tighter still: 90% of software development professionals had adopted AI tools at work by 2025, a 14 percentage-point year-over-year jump, per the DORA 2025 State of AI-Assisted Software Development (October 2025). DORA also found that 65% of respondents are heavily reliant on AI, and that developers spend a median two hours per day on AI-assisted work. Two hours per day is not supplemental — it is structural workflow integration.

The highest number in the dataset comes from DX: 91% AI adoption rate across 85,350 developers at 435 companies, per the DX Q4 2025 AI-Assisted Engineering Impact Report. DX's sample is not a random developer survey — it draws from organizations already using DX's engineering intelligence platform, which skews toward teams that have made explicit investments in developer tooling and are further along the adoption curve than the average Stack Overflow respondent.

One additional data point from GitHub: 80% of new developers on GitHub use Copilot within their first week, per the GitHub Octoverse 2025 (October 2025). For new entrants to the profession, AI-assisted coding is not an advanced technique — it is the default starting posture.

The tool-preference data has a structural problem: Stack Overflow and JetBrains measure adoption in fundamentally different ways, and their numbers for the same tools can diverge by 28–39 percentage points. The divergence table in the next section explains why. Here are the raw numbers from each survey, presented without reconciliation.

Stack Overflow 2025 — general AI tool usage (any context, not just coding): ChatGPT leads at 82% of developers, OpenAI GPT models at 81–82%, GitHub Copilot at 68%, Anthropic Claude Sonnet at 45% among professionals and 30% among learners. Cursor entered Stack Overflow's survey for the first time in 2025 at 18%. Claude Code debuted at 10%. Windsurf reached 5%. OpenAI Codex CLI, Google Antigravity, and JetBrains Junie were not surveyed by Stack Overflow.

JetBrains Jan 2026 — work adoption of specialized AI coding tools: The JetBrains "Which AI Coding Tools Do Developers Actually Use At Work?" (January 2026 sample, April 2026 publication) paints a different picture. GitHub Copilot has 76% awareness but only 29% global work adoption — rising to 40% at companies with 5,000+ employees. Cursor shows 69% awareness and 18% work adoption globally. Claude Code's trajectory is the standout: awareness climbed from 31% (April–June 2025) to 49% (September 2025) to 57% (January 2026), and work adoption rose from roughly 3% to 12% to 18% globally in the same period — reaching 24% in the US and Canada.

Tools not covered in Stack Overflow's 2025 survey but captured by JetBrains include: OpenAI Codex CLI at 27% awareness and 3% work adoption; Google Antigravity at 6% work adoption; JetBrains AI Assistant and Junie combined at 11% work adoption (AI Assistant 9%, Junie 5%). These omissions in Stack Overflow's tool list are not random — Stack Overflow's survey was fielded in mid-2025, before several of these tools had reached meaningful market presence.

No other compilation of AI coding statistics surfaces the cross-source contradictions explicitly. The table below assembles six metrics where multiple surveys ask the same (or very similar) question, shows each survey's answer side by side, and explains the spread. Understanding why surveys disagree is more useful than averaging them — the spread is itself a signal about how hard AI coding adoption is to measure.

The table below presents six divergence points across surveys. The "Spread" column is the gap between the highest and lowest reported value; where a survey did not ask the question, the cell is marked n/a.

The Copilot divergence deserves extended analysis because it is the largest and the most consequential for budget decisions. Stack Overflow's 68% reflects any developer who uses Copilot for any purpose — including casual or personal use outside of work. JetBrains' 29% measures work adoption specifically among developers using dedicated AI coding tools, in organizations where tool choices are deliberate rather than individual. The 40% figure at companies with 5,000+ employees aligns more closely with what enterprise software buyers report in their own usage audits. The practical takeaway: if you are benchmarking Copilot enterprise penetration, 40% is a more accurate reference point than 68%.

The Claude Code divergence is simpler: it is a time-gap problem. Stack Overflow's survey captured Claude Code at 10% in mid-2025. JetBrains captured it at 18% globally and 24% in the US and Canada six months later. Claude Code was not widely available in early 2025 — it launched with broader access in late 2024 and accelerated into 2025. The Claude Code 1.3 deep-dive covers the product evolution that drove this adoption arc.

Productivity data from DX's large-scale Q4 2025 survey provides the most granular picture of what daily AI use actually delivers in engineering throughput terms. The numbers are large enough to matter and specific enough to be useful for ROI models.

Average time saved. DX measured 3.6 hours per week of average time saved per developer across its tracked companies. Daily AI users save 4.1 hours per week; Staff+ engineers using AI daily save 4.4 hours per week, per the DX Q4 2025 Impact Report. The seniority gradient — more senior engineers save more time — suggests that AI tools scale with existing expertise rather than replacing it, which aligns with the DORA finding that AI functions as "an amplifier of existing strengths and weaknesses."

PR throughput. Daily AI users at DX-tracked companies merge 2.3 PRs per week (median) versus 1.4 PRs per week for non-users — a 60% throughput advantage. This is not a self-reported figure; it is measured from actual merged PR data. Engineering managers using AI daily ship twice as many PRs as rare or non-users.

Onboarding acceleration. DX found that time to 10th merged PR dropped from 91 days (no AI use) to 49 days (daily AI use) — new developer onboarding cut nearly in half. For engineering leaders, this is the single most compelling productivity argument for aggressive AI adoption: the ramp from hire to productive contributor is measurably shorter.

GitHub/Accenture RCT results. A randomized controlled trial across 4,800 Accenture developers (September 2024, the latest available large-scale RCT) found pull request cycle time dropped from 9.6 days to 2.4 days — a 75% reduction. The same study measured 8.69% more PRs per developer, 11% higher merge rate, and a 84% increase in successful builds. RCT data is methodologically stronger than survey self-reports; these figures are the closest thing to causal evidence in the dataset.

Broader survey figures. JetBrains found roughly 89% of developers save at least one hour per week using AI, with 20% (1 in 5) saving 8 or more hours per week. Stack Overflow found 51.6% of respondents reporting a positive productivity impact (16.3% to a great extent, 35.3% somewhat). The DORA 2025 report found over 80% of respondents say AI has enhanced their productivity — the strongest positive reading in the dataset.

The pattern across these sources is consistent: productivity gains are real and measurable, but concentrated among daily users and senior engineers. Occasional AI use produces marginal gains. Daily integrated use produces structural throughput improvements. The implication for team leads: the adoption rate is less important than the depth of use. For a full ROI framework, see the Claude Code ROI calculator and the broader AI agent productivity statistics roundup.

Code quality data is where the AI coding story gets most complicated. DORA and DX report positive outcomes; GitClear's objective code analysis points to measurable degradation in several quality indicators. Both can be true simultaneously — they are measuring different things.

The GitClear findings. GitClear analyzed 211 million lines of code across 2021–2024 and found four concerning trends. First, copy/pasted lines climbed from 8.3% (2021) to 12.3% (2024) — a 48% relative increase. Second, refactored ("moved") code lines fell from roughly 24–25% (2021) to under 10% (2024) — developers are writing more net-new code and refactoring less. Third, duplicate code-block frequency rose approximately 8x year-over-year in 2024 — and copy/paste lines exceeded refactored lines for the first time in GitClear's dataset. Fourth, code revised within two weeks of its initial commit grew from 3.1% (2020) to 5.7% (2024) — premature commits nearly doubled, suggesting AI-generated code reaches the repository before it is fully validated.

What developers report subjectively. Stack Overflow 2025 found 66% of developers cite "AI solutions that are almost right, but not quite" as a top frustration, and 45.2% say debugging AI-generated code is more time-consuming than writing it themselves. These self-reports are consistent with GitClear's objective data: AI generates code faster, but the output often requires substantial post-generation cleanup.

What organizations report at scale. DORA found 59% of respondents report a positive influence of AI on code quality. DX found that AI-authored code makes up 22% of merged code at the median company tracked, with daily AI users pushing 24% AI-authored merged code and monthly users at just over 20%. DX also found that increasing GenAI enablement by 25% correlates with roughly 8% higher code maintainability, 10.6% higher change confidence, 16.1% lower knowledge gaps, and 18.2% less time loss.

The reconciliation: DORA and DX capture developer and organizational perception, and they find that AI tools make developers feel more confident and produce more code. GitClear captures what that code actually looks like in the repository — more copy/paste, less refactoring, more churn within two weeks. The two sets of findings are not contradictory: AI-assisted coding may simultaneously increase developer confidence and decrease objective code maintainability. Teams that optimize for velocity without measuring churn will see the DORA numbers but not the GitClear signal until it shows up in technical debt accumulation 12–24 months later. The vibe coding anti-patterns guide covers the specific failure modes in detail.

The security picture in AI-generated code is worse than most development teams realize. The data below comes from three independent sources — Veracode's controlled benchmark, Apiiro's production monitoring data, and GitHub's Octoverse security metrics — and they point in the same direction: AI-assisted development introduces security vulnerabilities faster than existing remediation processes can absorb them.

Veracode 2025 GenAI Code Security Report. Veracode tested 80 coding tasks across 100+ LLMs (October 2025) and found that 45% of AI-generated code contains a security vulnerability across Java, JavaScript, Python, and C#. The variance by language is striking: Java had a 72% security failure rate — the riskiest language in the dataset. XSS-resistant secure code was generated in only 12–13% of relevant tasks. The "100+" framing is exact — Veracode tested more than 100 LLMs and did not specify the precise number.

Apiiro production findings. Apiiro's "4x velocity, 10x vulnerabilities" report (September 2025) monitored 7,000+ developers across 62,000 repositories and found that AI-assisted code triggered 10,000+ new security findings per month by June 2025 — a 10x increase in six months from December 2024 to June 2025. The same study found AI-assisted developers expose cloud credentials and keys (Azure Service Principals, Storage Access Keys) at nearly 2x the rate of non-AI peers. The dates matter: this is not year-over-year comparison — it is a six-month acceleration.

The credential-exposure finding deserves particular attention. Cloud credential exposure in version control is one of the highest-severity security incidents a development team can trigger — the attack window between commit and rotation is often measured in minutes. Doubling the exposure rate for AI-assisted developers is a finding that should change how teams configure their pre-commit hooks and secrets scanning.

The trust gap. Stack Overflow 2025 found trust in AI accuracy dropped to 29% in 2025 — down 11 percentage points from 40% in 2024. And 46% of developers actively distrust AI output (up from 31% the prior year). Despite this distrust, the Apiiro and Veracode data shows that security vulnerabilities are still reaching production — meaning developers are not applying the skepticism they report having.

Positive signals in the GitHub data. Not all the security news is negative. The GitHub Octoverse 2025 reports that GitHub's security tooling (Dependabot, code scanning, secret scanning) produced 30% faster critical-vulnerability fix time — median dropped from 37 to 26 days — and 26% fewer repositories with critical alerts year-over-year. Supply-chain risk also rose: Endor Labs research found that 92% of all npm maintainer account takeovers happened in 2025 alone, with a 14x increase in malware advisories versus two years prior.

Our AI transformation advisory consistently raises security governance as the first constraint to establish before scaling AI coding adoption. The data above is why: the velocity gains are real, but so is the security exposure, and they arrive simultaneously.

The enterprise and indie adoption patterns diverge in interesting ways. Large enterprises show higher Copilot penetration but lower overall adoption depth. Smaller companies show higher weekly AI use frequency. The tool economics also separate sharply by company size — Cursor's $2B ARR trajectory and 67% Fortune 500 penetration tells a different story than its 18% global work adoption number.

Copilot and enterprise concentration. GitHub reports 90% of the Fortune 100 use GitHub Copilot. Copilot reached approximately 20 million cumulative users by July 2025 and 4.7 million paid subscribers by January 2026 (a 75% year-over-year increase). Copilot Enterprise customers grew 75% quarter-over-quarter in Q2 2025. JetBrains' Jan 2026 data shows Copilot work adoption at 29% globally rising to 40% at companies with 5,000+ employees — confirming the enterprise concentration.

Cursor's enterprise penetration. Cursor reports 67% of Fortune 500 companies use it, with 7 million+ monthly active users and 1 million+ daily active users. ARR grew from $100 million (January 2025) to $2 billion (February 2026) — an approximately 20x trajectory in 13 months. The Fortune 500 penetration figure and the global 18% work adoption figure from JetBrains are not contradictory: Fortune 500 companies often evaluate and seat licenses for many employees without reaching majority daily use.

Claude Code's enterprise footprint. Anthropic reports 8 of the Fortune 10 are Claude customers. The company has 300,000+ business customers, with 500+ at $1 million or more in annualized spend. Claude Code's ARR run-rate went from $500 million (September 2025) to $2.5 billion (February 2026) — per the Anthropic Series G announcement.

Adoption rate by company size. DX's data provides the most granular company-size breakdown. Companies under 50 developers hit 80% weekly or higher AI usage frequency; companies with 50–200 developers hit just over 70%. Adoption depth decays with organizational size — larger organizations have more procurement friction, more security review requirements, and more heterogeneous tooling environments.

McKinsey context. The McKinsey State of AI 2025 found 88% of organizations now use AI in at least one business function (up from 78% in 2024), with software engineering reporting 10–20% cost reductions tied to AI — but only roughly 5.5% of organizations see real financial returns from AI. The gap between reported cost reduction and actual financial returns reflects the measurement problem: most organizations measure AI adoption, not AI value creation. GitHub's autonomous agent data adds another dimension: GitHub coding agents opened 1+ million PRs between May and September 2025, reaching approximately 1.2 million autonomous PRs per month at run-rate. And 50% of open source maintainers now use Copilot.

The regional data in AI coding adoption splits in two directions: developer population growth (where India dominates) and AI trust sentiment (where India also leads, and Germany trails every major market). These are not coincidental — emerging developer markets have lower legacy skepticism and more to gain from AI-assisted skill development.

India's developer growth. GitHub Octoverse 2025 reports India added 5.2 million or more new developers in 2025 — approximately 14% of GitHub's 36.2 million new signups for the year. India's developer population grew from 4.5 million (2020) to 21.9 million (2025) and is projected to reach 57.5 million by 2030, overtaking the US (projected at 54.7 million). Brazil has 6.89 million developers today, projected to reach 19.6 million by 2030 (third-largest community globally); China is projected at 17.7 million by 2030.

Language ecosystem shifts. GitHub Octoverse 2025 found TypeScript took the top position in contributors for the first time (2.6 million contributors, up 66% year-over-year). Python reached 2.6 million contributors (up 48% year-over-year). JavaScript: 2.15 million contributors (up 24.79% year-over-year). AI project repos on Python totaled 582,196 — up 50.7% year-over-year. The language ecosystem is tilting toward TypeScript and Python precisely because both are the dominant languages in AI tooling and AI application development.

Trust sentiment by country. Stack Overflow 2025 measured combined AI trust scores by country. India leads at 56% combined trust. Ukraine is at 41%. Italy 31%. Netherlands and the US at 28%. Poland 26%. Canada and France at 25%. UK at 23%. Germany is the lowest of the major developer markets at 22% — a 34 percentage-point gap versus India. Germany's lower trust aligns with its stronger regulatory culture and historical skepticism toward data-intensive technology products.

Career-stage adoption. Stack Overflow 2025 shows a monotonic relationship between career stage and daily AI use: early-career developers use AI daily at 55.5%, mid-career at 52.8%, experienced at 47.3%, and learners at 39.5%. The inverse intuition — that experienced developers would use AI more because they have the context to evaluate output — does not hold. New developers adopt AI coding tools as default workflow; experienced developers are more selective.

Microsoft Work Trend Index 2026. The Microsoft 2026 Annual Work Trend Index (April 2026) found that 66% of AI users say AI lets them spend more time on high-value work, and 58% are producing work they could not have produced a year ago — rising to 80% among "frontier professionals" (the highest-usage cohort). Additionally, 49% of all Microsoft 365 Copilot conversations now support cognitive work — analysis, decision-making, and creative thinking — rather than simple task execution. The Microsoft Work Trend Index figures were retrieved from secondary coverage summarizing the official report (Inc., theneuron.ai), with the canonical source cited as the Microsoft news page.

The original analytical inference here: the trust gap between India (56%) and Germany (22%) is not just a cultural curiosity — it predicts the next five years of global AI coding adoption geography. Developer populations growing fastest (India, Brazil) will adopt fastest and with the least institutional friction. Developer populations with the strongest data-privacy regulatory environments (Germany, UK) will adopt more slowly and with more compliance overhead. For global teams evaluating rollout sequencing, this geography matters for change management planning. Our AI transformation services include regional readiness assessments for exactly this reason.