AI in software and tech: what works and how to start
How many are using it
of US information businesses used AI in the fortnight 21 Sep 2026 to 4 Oct 2026, by the US Census Bureau’s Business Trends and Outlook Survey (published 22 Oct 2026).
Businesses, not workers: a firm counts if it used AI at all. Every sector · Over time
What works here
All rated use cases →Not enough evidence yet to say whether AI coding assistants make developers faster.
6 studies counted →Software engineeringDoes AI make code review faster or better?Not enough evidence yet to say whether AI makes code review faster or better.
4 studies counted →Research and analysisDoes AI help people analyse data more accurately?Mixed results: studies of AI for data analysis disagree on whether it helps.
7 studies counted →IT supportDoes AI help IT and security staff resolve problems faster?Not enough evidence yet to say whether AI helps IT and security staff resolve problems faster.
3 studies counted →What hasn't worked
Across all sectors →- Coding with AI, Becker et al. (2025): +19% Implementation time per issue (developer-logged time, before and after PR review), regression-adjusted for forecast difficulty.
- Coding with AI, He et al. (2026): +30.26% Static analysis warnings (SonarQube), average post-adoption effect.
- Coding with AI, He et al. (2026): +41.64% Code (cognitive) complexity, average post-adoption effect.
- Coding with AI, Agarwal et al. (2026): +17.73% Static analysis warnings, agent-first repositories (IDE-first estimate +19.00% is not significant).
- Coding with AI, Agarwal et al. (2026): +39% Cognitive complexity, both groups.
- Coding with AI, Agarwal et al. (2026): Repositories already using AI IDEs before agents (IDE-first): Only a short bump in commits (+16-28% in months 0-2), then lines added about -61% and commits about -35% by month 6, while complexity still rises (section 4)
- AI in code review, Maddila et al. (2025): +5.5% TimeInReview: time a diff spent in review when AI patches were shown to reviewers (Expt. 1).
- AI in code review, Cihan et al. (2024): +42% Average pull request closure time, before vs after the bot (5h52m to 8h20m; % computed).
- AI in code review, Cihan et al. (2024): Project #1: Closure time rose from 2h48m to 4h38m (p < 0.001)
- AI in code review, Cihan et al. (2024): Project #3: closure time rose from 20h22m to 30h51m (p < 0.001)
- Data analysis with AI, Fitzpatrick et al. (2025): −12% Accuracy of answers to six questions on insurer reporting templates, 40 values in all, graded against ground truth (32.3 Gen AI vs 36.5 control out of 40).
- Data analysis with AI, Fitzpatrick et al. (2025): Participants scoring under 100% on the baseline assessment (N=18): With Gen AI, the correct-answer rate fell from 79.63% on the assessment to 65.83% on the experiment task
- Data analysis with AI, Fitzpatrick et al. (2025): Participants scoring 100% on the baseline assessment (N=49): With Gen AI, the correct-answer rate fell from 100% to 84.48%; the baseline task was too easy to separate people
- Data analysis with AI, Brodeur et al. (2026): −0.95 major errors detected per team Major coding errors detected per team, AI-assisted vs human-only (0.74 vs 1.70).
- Data analysis with AI, Brodeur et al. (2026): Teams where ChatGPT led the work (minimal human oversight): Reproduced 56.8 points fewer papers than human-only teams (37%) and took about 119% longer.
Rules to check first
Source code and customer data sent to an AI tool leave your systems. Check the vendor's terms on storage and training, and your customers' contracts.
Treat AI-written code and suggested fixes as untrusted: review them as you would code from a new colleague.
Most internal coding and analysis help is not high-risk. Building products on general-purpose AI models brings duties for model providers and for high-risk uses. What applies when →
Before you start
Will AI help with your work? →- Measure, don't guessIn a 2025 trial, experienced developers expected AI to speed them up and were 19% slower. METR thinks newer tools help more, but calls its follow-up data very weak evidence, so time your own work.
- Keep people in chargeTeams that let the AI lead a data analysis reproduced far fewer results than teams that kept control.
- Watch quality, not just outputIn two studies of open-source projects, adopting AI coding tools meant more code, but also more warnings and more complex code.
- Place suggestions carefullyShowing AI fixes to reviewers slowed reviews; showing them to authors did not.
- Ask for independent evidenceEvery study of AI in IT support so far was run by the company selling the tool.
Learn for free
All free courses →More data: AI Adoption
More charts, each checked against its source, are on AI Adoption statistics. Dated events are in The AI Record.
How this page is made
The adoption figure is the US Census Bureau’s own estimate, read from its download; the use cases are rated from published studies by fixed rules (pilot, rules v0.3). Editorial owner: AI Adoption editors. Disclosure: AI Adoption uses AI tools, including Claude from Anthropic, in its research and editorial work. Spot a mistake? Corrections.