AI mobile app development statistics 2026: 56 primary-source facts

56 primary-source AI mobile app development statistics on adoption, productivity, trust, agents, AI projects, and mobile platform quality.

AppX team ·

Updated September 13, 2026. This page collects 56 primary-source facts about AI-assisted software development and the mobile app ecosystem. Every number links to the organization that produced the underlying survey, experiment, platform telemetry, or policy report.

Key takeaways

The clearest signal is broad adoption paired with unresolved trust, productivity, and quality questions.

  1. 84% of respondents use or plan to use AI tools in software development. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=33,662 question respondents; Use and intent are combined; this is not a representative census.)
  2. 46% actively distrust the accuracy of AI output. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=33,244 question respondents; Trust is self-reported and is not an observed error rate.)
  3. 52% say AI tools or agents had a positive effect on their productivity in the previous year. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=31,636 question respondents; This is perceived productivity, not elapsed-time measurement.)
  4. 87% report concerns about AI-agent accuracy. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=28,930 question respondents; Concern is not an observed failure rate.)
  5. 90% of respondents use AI at work. (Google DORA: State of AI-assisted Software Development 2025, 2025-09-23; Nearly 5,000 technology professionals; Self-reported use in DORA's survey sample.)
  6. AI-allowed tasks took 19% longer in this randomized trial. (METR: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, 2025-07-10; 246 issues completed by 16 experienced open-source developers; This narrow early-2025 setting does not estimate every developer or task.)
  7. GitHub counted more than 4.3 million AI-related repositories. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Repositories GitHub classified as AI-related; Platform classification and telemetry; activity does not prove production use.)
  8. More than 1.13 million public repositories imported a generative-AI SDK. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Public repositories on GitHub; Imports show a dependency in a repository, not active end-user adoption.)
  9. The App Store averaged 929,754,061 app downloads per week. (Apple: 2025 App Store Transparency Report, 2026; exact release date not stated; App Store downloads; Downloads are not necessarily first installs or unique people.)
  10. Android ran on more than 3 billion active devices across 190+ countries. (Google: The Android Show: I/O Edition 2025, 2025-05-13; Active Android devices; Devices are not people or phones exclusively.)

How this list was built

We included primary sources that expose the underlying figure and enough context to interpret it: Stack Overflow's 2025 survey, Google DORA's 2025 survey, METR's randomized trial, GitHub AI-project telemetry, Apple and Google platform reports, Android documentation, and Expo's own ecosystem counter. The figures were last checked on September 13, 2026.

The evidence types are not interchangeable. Surveys measure reported use, attitudes, or perceived outcomes. METR measured completion time in a narrow controlled setting. GitHub, Apple, and Google report activity on their own platforms. Policy thresholds describe a quality boundary, not the average app. Where a producer did not publish an exact release date or methodology for a live counter, we say so.

We exclude market forecasts, secondary statistics roundups, generic software activity, mobile-payment and connectivity figures, unsupported failure-rate claims, and vendor revenue used as a proxy for customer success. We will correct a figure when its primary source changes materially and review source links monthly. A year change will require substantive new data rather than a date-only refresh.

How widely are developers using AI?

AI use is now common in the surveyed developer populations, while agent use and vibe coding remain less universal.

  1. 60% had a favorable attitude toward AI tools in development. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=33,412 question respondents; Attitude is self-reported, not a measured outcome.)
  2. 72% said vibe coding was not part of their professional workflow. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=26,564 question respondents; The survey defined vibe coding as generating software from LLM prompts.)
  3. 52% either did not use AI agents or used simpler AI tools without agent behavior. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=31,877 question respondents; The survey defined agents as autonomous or requiring minimal intervention.)
  4. 38% had no plans to adopt AI agents. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=31,877 question respondents; This measures stated future intent.)
  5. 65% reported heavy reliance on AI for software development. (Google DORA: State of AI-assisted Software Development 2025, 2025-09-23; Nearly 5,000 technology professionals; This combines 37% moderate, 20% a lot, and 8% a great deal of reliance.)
  6. More than 80% said AI enhanced their productivity. (Google DORA: State of AI-assisted Software Development 2025, 2025-09-23; Nearly 5,000 technology professionals; Self-reported productivity; no elapsed-time claim.)
  7. 59% reported a positive influence on code quality. (Google DORA: State of AI-assisted Software Development 2025, 2025-09-23; Nearly 5,000 technology professionals; Perception, not an independent defect-rate measurement.)

Do developers trust AI-generated code?

Many developers use AI while checking its output carefully and keeping people in high-risk parts of the workflow.

  1. 33% actively trusted the accuracy of AI output. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=33,244 question respondents; Trust is self-reported and scale-specific.)
  2. Only 3% highly trusted the accuracy of AI output. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=33,244 question respondents; This is not a measured code-quality rate.)
  3. 66% cited AI solutions that are almost right, but not quite, as a frustration. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=31,476 question respondents; Multi-select self-report; choices are not mutually exclusive.)
  4. 45% said debugging AI-generated code takes more time. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=31,476 question respondents; Perceived burden, not a timed study.)
  5. 75% would ask a person when they did not trust an AI answer. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=29,163 question respondents; Response to a hypothetical scenario.)
  6. 76% did not plan to use AI for deployment and monitoring. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=11,202 workflow-question respondents; A three-to-five-year intention item, not current behavior.)
  7. 69% did not plan to use AI for project planning. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=11,202 workflow-question respondents; A three-to-five-year intention item, not current behavior.)
  8. 81% reported security or privacy concerns about AI agents. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=28,930 question respondents; Concern is not an observed incident rate.)
  9. 30% reported little or no trust in AI. (Google DORA: State of AI-assisted Software Development 2025, 2025-09-23; Nearly 5,000 technology professionals; This combines 23% little trust and 7% no trust; it is not an error rate.)

How are developers using AI agents?

Developers who already use agents mostly apply them to software work and often report time savings, while team-level gains are less common.

  1. 84% of software developers who used agents at work used them for software development. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=12,301 relevant question respondents; Conditional on already using agents at work.)
  2. 70% of agent users said agents reduced time on specific development tasks. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=12,823 agent-impact respondents; Agreement among agent users, not a controlled time study.)
  3. 69% of agent users said agents increased productivity. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=12,823 agent-impact respondents; Agreement among agent users, not an objective output measure.)
  4. Only 17% of agent users said agents improved team collaboration. (Stack Overflow: 2025 Developer Survey: AI, 2025-07-29; n=12,823 agent-impact respondents; Agreement among agent users; team structures vary.)

Does AI actually make experienced developers faster?

The strongest controlled result in this source set shows that perceived speed and measured completion time can move in opposite directions.

  1. Before starting the tasks, developers forecast that AI would reduce completion time by 24%. (METR: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, 2025-07-10; 16 experienced open-source developers across 246 randomized issues; A forecast, not the observed result.)
  2. After the work, developers still estimated that AI had reduced their completion time by 20%. (METR: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, 2025-07-10; 16 experienced open-source developers across completed issues; A post-task perception that differed from measured time.)
  3. 75% of the developers experienced a slowdown with AI allowed. (METR: Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, 2025-07-10; 12 of 16 experienced open-source developers; Tiny purposive sample; do not interpret as population prevalence.)

How much AI-related software is being built?

GitHub telemetry shows rapid growth in AI repositories and agent activity, but the platform warns that these observations do not establish that AI caused broader activity growth.

  1. More than 693,000 public repositories that import a generative-AI SDK were created in the preceding 12 months. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Public repositories importing selected generative-AI SDKs; Repository dependency use does not prove production deployment.)
  2. Public repositories importing a generative-AI SDK grew 178% year over year. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Public repositories importing selected generative-AI SDKs; GitHub comparison for 2025-08 versus 2024-08.)
  3. Roughly 80% of new GitHub users tried Copilot in their first week. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; New GitHub users; GitHub product telemetry; trying is not sustained use or success.)
  4. Generative-AI projects averaged about 151,000 monthly contributors. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Distinct monthly contributors to GitHub generative-AI projects; GitHub classification and platform telemetry.)
  5. Generative-AI project contributors peaked at 206,830 in May 2025. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Distinct monthly contributors to GitHub generative-AI projects; Peak month in GitHub's measurement year.)
  6. From January through August 2025, generative-AI projects averaged about 175,000 monthly contributors, up 108% year over year. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Distinct monthly contributors to GitHub generative-AI projects; Like-for-like platform comparison, not a census of all AI developers.)
  7. Monthly contributions to AI projects reached about 6 million in August 2025. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Contribution events in GitHub AI-related projects; Activity volume does not show app quality or revenue.)
  8. 50% of open-source projects had at least one maintainer using GitHub Copilot. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Open-source projects in GitHub's report; Use by one maintainer does not imply project-wide adoption.)
  9. More than 1 million pull requests were created by Copilot coding agent from May through September 2025. (GitHub: Octoverse 2025, 2025-10-28; updated 2026-02-28; Copilot coding agent pull requests; Product telemetry; GitHub reports strong repository selection effects.)

How large is the mobile app market and its quality bar?

Mobile distribution reaches billions of devices and hundreds of millions of weekly downloads, while store review and reliability thresholds remain substantial constraints.

  1. The App Store contained 2,172,472 apps. (Apple: 2025 App Store Transparency Report, 2026; exact release date not stated; Apps available in the App Store; Platform-reported count.)
  2. Apple reviewed 9,100,620 app submissions. (Apple: 2025 App Store Transparency Report, 2026; exact release date not stated; App Store submissions reviewed; One app can be submitted repeatedly.)
  3. Apple rejected 2,093,244 app submissions. (Apple: 2025 App Store Transparency Report, 2026; exact release date not stated; App Store submissions rejected; A submission may violate more than one rule; this is not unique apps.)
  4. 1,354,418 submissions were rejected under Apple's Performance category. (Apple: 2025 App Store Transparency Report, 2026; exact release date not stated; App Store submission rejections; Category count, not unique apps; rejection categories can overlap.)
  5. The App Store averaged 892,670,823 visitors per week. (Apple: 2025 App Store Transparency Report, 2026; exact release date not stated; App Store visitors; Weekly average, not unique people across the year.)
  6. An average of 463,575,966 customer accounts searched the App Store each week. (Apple: 2025 App Store Transparency Report, 2026; exact release date not stated; App Store customer accounts performing search; Accounts are not necessarily unique people.)
  7. Google prevented 2.36 million policy-violating apps from being published on Play. (Google: How we kept Google Play and the Android app ecosystem safe in 2024, 2025-01-29; App submissions reviewed by Google Play; Google's internal prevention count, not the Play catalog size.)
  8. Google banned more than 158,000 bad developer accounts. (Google: How we kept Google Play and the Android app ecosystem safe in 2024, 2025-01-29; Google Play developer accounts; Accounts Google classified as harmful.)
  9. Google prevented 1.3 million apps from gaining unnecessary or excessive access to sensitive user data. (Google: How we kept Google Play and the Android app ecosystem safe in 2024, 2025-01-29; Apps reviewed under Google Play policy; A policy-enforcement measurement.)
  10. More than 92% of Google's human reviews for harmful apps were AI-assisted. (Google: How we kept Google Play and the Android app ecosystem safe in 2024, 2025-01-29; Google human reviews for harmful apps; Narrow internal review workflow; it is not the share of all app reviews.)
  11. More than 91% of Play app installs used Android 13 or newer protections. (Google: How we kept Google Play and the Android app ecosystem safe in 2024, 2025-01-29; Google Play app installs; Installs are not active devices or people.)
  12. Google Play's overall bad-behavior threshold for user-perceived crash rate is 1.09% of daily active users. (Android Developers: Technical quality: Android vitals, Current documentation; retrieved 2026-09-13; Daily active users for an Android app; A policy threshold, not an industry-average crash rate.)
  13. Google Play's overall bad-behavior threshold for user-perceived ANR rate is 0.47% of daily active users. (Android Developers: Technical quality: Android vitals, Current documentation; retrieved 2026-09-13; Daily active users for an Android app; A policy threshold, not an industry-average ANR rate.)
  14. Expo reported more than 100,000 weekly active developers. (Expo: About Expo, Undated live page; retrieved 2026-09-13; Expo weekly active developers; Rounded, operator-reported live counter with no published methodology.)

What should an app team do with these numbers?

Use them as boundary conditions, not promises. High AI adoption does not guarantee faster delivery. Repository growth does not prove that a product reached users. A phone preview does not prove that an app is ready for production or store review.

Start with one bounded user journey, make the first version testable, and record what the AI tool produced, what a person corrected, and what the user could complete. Our app-prompt template helps define that journey. Then test the result on a phone with Expo Go and use our source-export handoff guide when a developer needs to continue the work.

Source and correction policy

Each statistic appears once, links to its primary producer, states the measured population or platform, and carries a short limitation. For every statistic, we recorded its publication date, measurement period, geography, retrieval date, and source-specific caveat. If you spot a primary-source update or a transcription error, contact AppX so we can verify and correct the page.

Try your own app idea

Describe your app in AppX →