Now accepting guest editorial from SAP, Oracle, Workday and ServiceNow practitioners.Pitch a guide
The Sign-Off Table
All statistics

Statistics · Published

AI Coding Assistant Statistics (2026)

In a 2025 randomized trial, experienced open-source developers took 19% longer with AI tools while believing they were faster.

Compiled by Matt Angerer
23 sources · 12 min read

AI coding assistants are now routine in most development teams, but the evidence on what they do to delivery speed, code quality and test effort is mixed. This roundup collects the figures we could source from surveys, field experiments, analyst forecasts and vendor disclosures published between 2022 and 2026, and flags where they disagree. Most of the numbers are general to software engineering, with SAP ABAP and quality-engineering data called out where they exist.

Key takeaways

79%

of developers reported using AI tools in 2025, up from 62% in 2024

Source: Stack Overflow, 2026 retrospective of the 2025 survey

19%

longer task completion with AI tools allowed, in a 2025 randomized trial of experienced developers

Source: METR, 2025

44%

of AI code generation tasks produced code with a known vulnerability, in Veracode's 2026 tests

Source: Veracode, 2026 GenAI Code Security Report

15%

of organizations had scaled generative AI across quality engineering enterprise-wide, in the 2025 World Quality Report

Source: Capgemini, World Quality Report 2025

Adoption: how many developers use AI tools

Adoption has climbed every year the surveys have asked, although the measures differ. Read the figures below as separate questions rather than one trend line.

  • Stack Overflow's 2025 survey found that 84% of respondents use AI tools or plan to, up from 76% a year earlier (84% saying they use or plan to use AI tools in their development process, up from 76% in 2024).
  • Stack Overflow's later retrospective reports actual use at 44% in 2023, 62% in 2024 and 79% in 2025, with almost half of all respondents indicating they used AI tools daily.
  • For professional developers only, daily use was 51% in the 2025 survey (51% of professional developers use AI tools daily). That figure and the "almost half" figure above come from different populations, so they should not be read as conflicting.
  • Google Cloud's 2025 DORA report, based on nearly 5,000 technology professionals, put AI use at 90% of respondents. Google's own write-up calls this a 14% increase from last year. The post does not say whether that means 14 percentage points or 14 percent, so treat the year-over-year change with caution.
84%
90%

Source: Stack Overflow, 2025 Developer Survey; Google Cloud, 2025 DORA report

Source: Stack Overflow, retrospective published October 2026. Share of respondents reporting AI tool use in the development workflow.

Trust and sentiment

Use is rising while confidence falls. The same surveys that show record adoption also show developers doubting what the tools produce.

  • Stack Overflow's 2025 data put trust in AI output accuracy at 29%, a figure down from much higher levels in prior surveys (reported by DevOps.com).
  • Favorable views of AI tools dropped from over 70% in prior years to around 60% in 2025.
  • The top frustration was output that is close but not right, with 64% citing 'almost right' AI output and debugging AI code as their top frustrations.
  • Only 38% of respondents said AI agents had improved the quality of their code (Only 38% said AI agents have improved the quality of their code.).
29%
64%
38%

Source: DevOps.com, on Stack Overflow 2025 survey; InfoWorld, 2025

Agents: the fast-growing category

Agentic tools, which plan and execute multistep work, are the fastest-moving part of the market, and they are where practitioners should expect the most change in the next 12 months.

  • Agent use was 31% in Stack Overflow's 2025 survey, and this time we see agent use as almost doubled (59%) in an April 2026 pulse survey.
  • Among agent users, Claude Code usage increasing in usage (from 41% to 55%), according to the same retrospective.
  • In the 2025 survey, 52% of developers either don't use them or stick to simpler AI tools, while 38% have no plans to adopt them.
59%
51%

Source: Stack Overflow, 2026 retrospective; Stack Overflow, 2025 Developer Survey

Productivity: the evidence conflicts

Productivity results depend heavily on the task, the codebase and whether the study measured speed, output or perception. The four most cited results point in different directions.

  • METR, 2025 (experienced open-source developers, real issues). Allowing AI increases completion time by 19%, meaning developers are slower when using AI tooling. The study used 16 developers and 246 real issues, and the developers forecast that AI would cut their completion time by 24 percent; even after finishing and experiencing the slowdown, they still estimated AI had sped them up by about 20 percent. The tools were mostly Cursor Pro with Claude 3.5 and 3.7 Sonnet, the frontier models of the moment. METR's own text also cites the result as a 20% slowdown in its 2026 update, so expect both figures in secondary coverage.
  • METR, 2026 follow-up. METR now says the 2025 finding is now outdated. Its newer data suggest late-2025 AI likely accelerated open-source developers, but selection effects obscure the true speedup. METR has since changed its experiment design, and it warns that self-reported speedups can be quite unreliable.
  • GitHub-published controlled experiment, 2022 (historical). A test with 95 developers, of whom 45 received Copilot, found the Copilot group finished faster on a defined task. The completion times were 71.17 minutes versus 160.89 minutes, a 55.8% difference, with a 95% confidence interval of 21% to 89%. This was one HTTP-server task, not a full delivery cycle.
  • Microsoft, 2026 observational analysis. A within-engineer analysis of 16,223 engineers found about 40% more pull requests in heavy-use weeks. Because it is observational, it shows association within each person's own weeks, not a controlled effect.
  • A field study of Copilot in an enterprise setting found that 94 participants rated GitHub Copilot's impact on overall productivity at an average of 3.95 out of 5, but GitHub Copilot adoption revealed no statistically significant improvement in engineering output metrics.
  • Gartner, in a 2025 forecast, expects "a 30% productivity gain in software development" across enterprises through 2028. That is a forecast, not a measured result.
Copilot task completion, 2022 controlled test (faster)55.8%
METR 2025, task time with AI allowed (longer)+19%

Source: Quash, GitHub Copilot statistics 2026, summarizing the 2022 GitHub experiment; METR, 2025. Bars show different measures and are not directly comparable: one is a faster-completion result, the other a slower-completion result.

Delivery stability and code security

Faster coding does not automatically mean safer releases. The DORA findings and Veracode's tests both point to downstream risk.

  • DORA's 2025 report found a positive relationship between AI adoption on both software delivery throughput and product performance.
  • In the same report, AI adoption does continue to have a negative relationship with software delivery stability.
  • DORA's 2025 data also showed that 65% of those surveyed are heavily relying on AI for software development, with 37% at a moderate level of reliance and 20% at "a lot."
  • Veracode's 2025 GenAI report found AI-generated code introduced risky security flaws in 45% of tests.
  • Its 2026 report put the figure at roughly 44% of tasks, with the average security pass rate across models is 56% – barely changed from 55% in the first report.
  • Syntax is no longer the weak point. Veracode says modern models now generate syntactically correct code nearly 100% of the time.
  • A later Veracode update shows a lower rate. It reports that Veracode finds AI-generated code security improved to 62%, yet 38% remains vulnerable. This conflicts with the 44% figure above. The two numbers come from different test snapshots and model sets, and we have not reconciled them.
  • Veracode's 2026 State of Software Security report found that 82 percent of organizations now carry security debt (vulnerabilities unresolved for more than a year).
Veracode 2025 report: tests with risky flaws45%
Veracode 2026 report: tasks with known vulnerability44%
Veracode later update: still vulnerable38%

Source: Veracode, October 2025 update; Veracode, 2026 GenAI Code Security Report; SecurityInformed, reporting a later Veracode update. Veracode sells application security testing, so these are vendor data.

Enterprise adoption and forecasts

Analyst forecasts have moved up quickly. Gartner's 2025 forecast is higher than its 2024 forecast for the same 2028 milestone, and the 2026 report shifts the frame to agents.

  • In 2024, Gartner forecast that by 2028, 75% of enterprise software engineers will use AI code assistants, up from less than 10% in early 2023. The same release reported that sixty-three percent of organizations are currently piloting, deploying or have already deployed AI code assistants in a Q3 2023 survey of 598 respondents.
  • In 2025, Gartner raised the 2028 figure to 90% of enterprise software engineers will use AI code assistants, up from less than 14% in early 2024.
  • Gartner's 2026 report on enterprise AI coding agents predicts that by 2028, the firm predicts that more than 70 percent of enterprise software engineers will rely on AI coding agents for both synchronous and asynchronous development tasks.
  • Gartner also expects that by 2027, more than 65 percent of engineering teams using agentic coding will treat IDEs as optional, shifting control, governance and validation to automated platforms.
Early 2023 (actual, per Gartner 2024)<10%
Early 2024 (actual, per Gartner 2025)<14%
2028 forecast, Gartner 202475%
2028 forecast, Gartner 202590%

Source: Gartner, April 2024; Gartner, July 2025. The 2023 and early-2024 bars are historical actuals; the 2028 bars are forecasts.

Testing and quality engineering

The most useful quality-engineering data for this audience comes from Capgemini's World Quality Report, which tracks testing practices across organizations. Its 2025 edition shows broad piloting and a small number of enterprise-scale deployments.

  • In the 2025 edition, 89% of responding organizations are piloting or deploying GenAI–augmented workflows, with 37% in production and 52% in pilot phases.
  • Only 15% have achieved enterprise-scale deployment, even though nearly 90% say they are pursuing it.
  • Capgemini's research page gives a different headline for the same question: 43% of organizations are experimenting with Gen AI in QA, but only 15% have scaled it enterprise-wide. The 89% figure counts piloting and deploying, and the 43% figure counts experimenting. The two are not the same measure.
  • Non-adoption is falling. Capgemini reports that the rate of non-adopters of GenAI increased to 11%, up from 4% in 2024, but still considerably lower than 2023's 31%. The 2025 rise is a reversal in direction, not a sign of retreat.
  • The report puts the productivity effect at an average productivity boost of 19%, but one third have seen minimal gains.
  • Test data is the constraint teams name most often. 60% of organizations struggle with secure, scalable test data, while 58% cite challenges in adopting AI-powered tools.
  • Synthetic data use has grown quickly. Capgemini says the use of synthetic data in testing has surged, rising from 14% in 2024 to an average of 25% in 2025.

Source: Capgemini, World Quality Report 2025. Share of organizations that are not adopting GenAI in quality engineering.

Organizations piloting or deploying GenAI in QE89%
In production37%
Enterprise-scale deployment15%

Source: Capgemini, World Quality Report 2025 press release, November 2025.

Data privacy risks67%
Integration complexity64%
Hallucination and reliability concerns60%
Lack AI/ML expertise (unchanged from 2024)50%

Source: Capgemini, World Quality Report 2025, via PR Newswire; skills figure from the Capgemini press release.

SAP and ABAP AI

SAP's own AI coding tools have published productivity claims. These are vendor data, and the figures below have not been independently replicated.

  • SAP's Q3 2025 release highlights state that with SAP Joule for Developers, ABAP developers can expect a 20%* reduction in time and effort to write ABAP code, a 25%* reduction in time and effort to test ABAP code. The asterisk is SAP's, and the release does not publish the test population.
  • SAP's July 2025 post cites an IDC white paper, with 80% of developers reported increased productivity when using AI coding assistants, with an average productivity gain of 35%. IDC's underlying sample is not described in the SAP post, so treat this as IDC data as relayed by SAP.
  • The most-used features, per the same post, were unit test generation, code generation, and autocompletion. For test managers, unit-test generation is the feature to validate first.
  • On availability, SAP said ABAP AI would come to S/4HANA Cloud Private Edition for all releases from 2021 onwards, with general availability planned for Q2 2026.
  • Pricing moved to a usage basis. As of Q2 2026, we are introducing a consumption-based pricing model for SAP Joule for Developers, ABAP AI capabilities, built on AI units.
Reduction in time to write ABAP code (SAP-stated)20%
Reduction in time to test ABAP code (SAP-stated)25%
Average productivity gain (IDC, as cited by SAP)35%

Source: SAP News Center, Q3 2025 release highlights; SAP News Center, July 2025. Vendor-published data.

What this means for your sign-off

  • Treat vendor productivity percentages as hypotheses. The SAP figures and the METR and GitHub results all measure different things. Before signing off on an AI-assisted ABAP change set, measure cycle time and defect escape on a sample of your own tickets, as the METR design did. Reuse a template such as the SAP Test Automation ROI Calculator to keep the comparison consistent.
  • Put review ahead of acceptance. The gap between speed and stability in DORA's data is the reason to tie AI-generated code to a documented review step. The AI-Generated Test Review Checklist and Log is one way to record that review.
  • Add security checks to AI-generated code paths. Veracode's findings on flaw rates justify running static analysis on any generated code that touches authorization, input handling or cryptography, regardless of whether the code compiles and passes unit tests.
  • Protect test data before you scale generation. Capgemini's respondents named test data as the top constraint, and synthetic data use is growing. The guide to test data for S/4HANA covers copies, masking and synthetic data options.
  • Govern agents separately from assistants. Agent use nearly doubled in a single year, and Gartner expects most teams to rely on agents within a few years. Agent output needs its own approval gate and audit trail, which is different from an autocomplete review. For SAP-specific agent work, see the practitioner guide to SAP AI.
  • Test AI-driven test automation the same way you test any tool. Generated test scripts can pass for the wrong reasons. The pitfalls of using Claude or Claude with Playwright for SAP testing covers the failure modes to check before a tool enters your regression suite.

Frequently asked questions

How many developers use AI coding tools?

In Stack Overflow's 2025 survey, 84% saying they use or plan to use AI tools in their development process, up from 76% in 2024. Its later retrospective puts actual use at 79% in 2025.

Do AI coding assistants make experienced developers faster?

The evidence conflicts. METR's 2025 trial found a 19% increase in completion time for experienced developers on their own repositories, while METR's 2026 follow-up says the earlier result is outdated and newer data suggest late-2025 AI likely accelerated open-source developers. A 2022 GitHub experiment on a single task found a 55.8% reduction in completion time.

How often does AI-generated code contain a known vulnerability?

Veracode's 2026 tests found that roughly 44% of tested AI code generation tasks resulted in code with a known vulnerability. A later Veracode update reports 38%. Veracode sells security testing, so treat both as vendor data.

Is SAP's ABAP AI productivity claim independently verified?

Not in the sources we reviewed. SAP states a 25%* reduction in time and effort to test ABAP code, and the asterisk refers to SAP's own footnote. Measure it on your own code before you rely on it.

How many organizations have scaled generative AI in quality engineering?

Capgemini's 2025 report found that nearly 90% of organizations are now actively pursuing generative AI (Gen AI) in their quality engineering (QE) practices, only 15% have achieved enterprise-scale deployment.

How widely will AI coding agents be used by 2028?

Gartner's 2026 forecast is that by 2028, the firm predicts that more than 70 percent of enterprise software engineers will rely on AI coding agents for both synchronous and asynchronous development tasks. This is a forecast, and Gartner's 2025 figure for AI code assistant use by 2028 was 90%.

Sources

  1. Stack Overflow's 2025 Developer Survey Reveals Trust in AI at an All Time Low, Stack Overflow, 2025
  2. Stack Overflow 2025 Developer Survey: AI section, Stack Overflow, 2025
  3. A look back before we look forward: A Developer Survey retrospective, Stack Overflow, October 2026
  4. AI use among software developers grows but trust remains an issue, InfoWorld, 2025
  5. Stack Overflow Survey Shows AI Adoption for Devs, DevOps.com, 2025
  6. Announcing the 2025 DORA Report: State of AI-Assisted Software Development, Google Cloud, September 2025
  7. How are developers using AI? Inside Google's 2025 DORA report, Google, 2025
  8. Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, METR, July 2025
  9. METR blog: early 2025 AI and experienced open-source developers, METR, July 2025
  10. We are Changing our Developer Productivity Experiment Design, METR, February 2026
  11. GitHub Copilot Statistic: What the Numbers Mean in 2026, Quash, 2026 (secondary summary of the 2022 GitHub experiment)
  12. GitHub Copilot and Developer Productivity: An Observational Dose-Response Analysis, arXiv, May 2026
  13. Beyond the Hype: The Efficiency-Throughput Gap with GitHub Copilot, Communications of the ACM, 2026
  14. Gartner Identifies the Top Strategic Trends in Software Engineering for 2025 and Beyond, Gartner, July 2025
  15. Gartner Says 75% of Enterprise Software Engineers Will Use AI Code Assistants by 2028, Gartner, April 2024
  16. AI Firms Push Cloud Giants from 'Leaders' Quadrant in Gartner AI Coding Report, Virtualization Review, June 2026 (secondary summary of Gartner's 2026 report)
  17. World Quality Report 2025: AI adoption surges in Quality Engineering, but enterprise-level scaling remains elusive, Capgemini, November 2025
  18. World Quality Report 2025-26, Capgemini, 2025
  19. 2026 GenAI Code Security Report: AI Is Writing More of Your Code but Security Hasn't Caught Up, Veracode, 2026
  20. Veracode October 2025 Update: GenAI Code Security Report, Veracode, October 2025
  21. Veracode Report Reveals Security Risks In AI Code, SecurityInformed, 2026 (secondary summary of a later Veracode update)
  22. SAP Business AI: Release Highlights Q3 2025, SAP News Center, October 2025
  23. How Joule for Developers and ABAP AI Capabilities Transform the Developer Experience, SAP News Center, July 2025

Last updated: November 16, 2026. Statistics are refreshed as new research is published.

More statistics

How this page is made: statistics are gathered from published research with AI-assisted search, every figure is linked to its source, and the page is refreshed as new data is published. Found an error? Tell us.