AI Coding Tools by the Numbers (2026)

A data analysis of AI coding tool adoption, the productivity evidence on both sides, which assistants developers actually use, and what the numbers say about choosing one.

AI coding tools are now near-universal and openly distrusted at the same time. 84% of developers use or plan to use them, GitHub Copilot passed 20 million users, and a controlled GitHub study found 55% faster task completion. Yet favorability is falling, and a 2025 randomized trial found experienced developers were 19% slower with AI on mature code. The numbers below show where AI coding tools help, where they hurt, and how to pick one.

Key takeaways

84% of developers use or plan to use AI tools, up from 76% a year earlier, per the 2025 Stack Overflow Developer Survey
GitHub Copilot crossed 20 million users, up 5 million in one quarter, per Microsoft’s Q4 FY25 earnings
A controlled GitHub study found 55% faster task completion with Copilot (GitHub research)
A METR randomized trial found the opposite — experienced developers were 19% slower with AI on mature codebases (METR, 2025)
Favorability fell to 60% from over 70% two years earlier; 46% of developers now distrust AI accuracy versus 33% who trust it (Stack Overflow 2025)
66% of developers’ top frustration is “AI solutions that are almost right, but not quite” (Stack Overflow 2025)
Most-used assistants: ChatGPT (81.7%), GitHub Copilot (67.9%), Google Gemini (47.4%), Claude Code (40.8%) (Stack Overflow 2025)

Headline numbers

Adoption is high; sentiment is not

Use is near-universal, but only 60% are favorable and 46% actively distrust accuracy. Stack Overflow 2025 Developer Survey.

Use or plan to use AI

84%

Favorable toward AI

60%

Use AI daily (pros)

51%

Distrust AI accuracy

46%

How many developers use AI coding tools?

84% of developers use or plan to use AI tools in their workflow, up from 76% the prior year, per the 2025 Stack Overflow Developer Survey. Among professional developers, 51% use them daily — AI assistance is now the default, not the exception.

Vendor adoption confirms the scale. GitHub Copilot crossed 20 million users as of Microsoft’s Q4 FY25 earnings, up more than 5 million in a single quarter, with over 90% of the Fortune 100 on board. CEO Thomas Dohmke confirmed the milestone directly.

The story underneath the headline is saturation. When 84% of developers already reach for these tools, the open question is no longer whether to adopt AI but which tool earns the seat — the exact decision an opinionated ai-coding shortlist exists to answer.

AI tool adoption

84 of every 100 developers use or plan to use AI

Each filled cell is one developer. Stack Overflow 2025 Developer Survey, up from 76 in 2024.

Do AI coding tools actually make developers faster?

The evidence points both ways, and the difference is the codebase. GitHub’s controlled study found developers completed a greenfield task 55% faster with Copilot — 1 hour 11 minutes versus 2 hours 41 minutes — a statistically significant result, per GitHub’s research.

A 2025 randomized trial by METR found the reverse. Experienced open-source maintainers were 19% slower with AI on large, mature projects they knew intimately, per METR. The perception gap was the striking part: those same developers believed AI had sped them up by 20%.

Both studies are credible; they measured different work. AI accelerates unfamiliar, greenfield, or boilerplate tasks and drags on deep work in a codebase the developer already masters.

The productivity split

Same tools, opposite results — because the task differs

Two rigorous studies. The gap is greenfield task versus mature codebase, not good tool versus bad tool.

Study	Setting	Result
GitHub (controlled)	Greenfield HTTP server task	55% faster
METR (RCT, 2025)	Mature open-source projects	19% slower
METR — perception	Same developers, self-estimate	Believed 20% faster

Which AI coding tools do developers use most?

ChatGPT leads general use, but purpose-built coding assistants dominate the IDE. Among developers using AI agents, ChatGPT (81.7%), GitHub Copilot (67.9%), Google Gemini (47.4%), and Claude Code (40.8%) are the most used, per the 2025 Stack Overflow Developer Survey.

The list mixes two categories. ChatGPT and Gemini are general assistants developers paste code into; GitHub Copilot and Claude Code live inside the editor and act on the repo. Both patterns are common on the same developer’s machine.

Most-used AI tools

What developers actually run

Share of AI-agent users, Stack Overflow 2025 Developer Survey. General assistants and in-editor tools coexist.

ChatGPT

81.7%

GitHub Copilot

67.9%

Google Gemini

47.4%

Claude Code

40.8%

Why is developer trust in AI falling?

Adoption rose while trust dropped. Favorability fell to 60% in 2025 from over 70% two years earlier, and 46% of developers now actively distrust AI accuracy against just 33% who trust it, per Stack Overflow. Only 3% say they highly trust the output.

The frustration is specific, not vague. 66% name “AI solutions that are almost right, but not quite” as their top complaint, and 45% say debugging AI-generated code takes longer than writing it themselves. The near-miss is worse than an obvious miss because it hides.

Trust is also uneven by workflow. Developers accept AI for searching answers (54%) but reject it for deployment and monitoring (76% won’t use it there) — the higher the blast radius, the lower the trust.

Where do AI coding tools help most, and least?

AI coding tools help most on unfamiliar and repetitive work and least on complex code the developer already knows. GitHub’s survey found 87% of developers preserved mental effort on repetitive tasks and 73% stayed in flow, per GitHub research.

The benefit skews toward less-experienced developers. GitHub’s controlled study found the largest gains among developers with less programming experience — the same population the METR trial shows AI can slow when the code is mature and the developer is expert.

Where AI clearly wins:

Boilerplate and scaffolding — new servers, config, test stubs, one-off scripts
Unfamiliar languages or APIs — the tool front-loads syntax you’d otherwise look up
Onboarding and learning — 44% of new coders lean on AI, per Stack Overflow

Where it drags:

Deep changes in a mature codebase you already know cold (the METR finding)
Debugging AI’s own near-misses — the 45% time sink
High-stakes deployment and monitoring — where 76% won’t trust it

What does this mean for choosing an AI coding tool?

Pick for the task, not the hype. The data says AI coding tools are worth the seat for greenfield work, unfamiliar stacks, and onboarding — and worth skipping for deep work in a codebase your team already masters. The 80/20 move is to standardize on one in-editor assistant and one general model, then measure your own throughput rather than trusting the vendor’s demo.

Practical selection rules from the numbers:

Match the tool to the work. In-editor assistants like GitHub Copilot and Cursor earn their keep on scaffolding; a general model handles one-off questions.
Distrust the near-miss. With 66% citing “almost right” as the top frustration, budget review time — AI-generated code is a draft, not a merge.
Measure, don’t assume. Developers in the METR study believed they were 20% faster while being 19% slower. Track your own cycle time.
Standardize to fight sprawl. Ten AI seats seeping in is a cost problem — see the real cost of SaaS sprawl.

That is why our ai-coding category scores tools on real, hands-on testing rather than adoption counts. See also how AI search changed software buying.

Frequently asked questions

What percentage of developers use AI coding tools in 2026?

84% use or plan to use AI tools, up from 76% a year earlier, per the 2025 Stack Overflow Developer Survey. Among professional developers, 51% use them daily. Adoption is effectively universal, even though favorability has fallen to 60%.

Do AI coding tools really make developers faster?

It depends on the task. A GitHub controlled study found 55% faster completion on a greenfield task, while a METR randomized trial found experienced developers 19% slower on mature code. AI helps most on new or unfamiliar work.

Which AI coding assistant is most popular?

Among AI-agent users, ChatGPT leads at 81.7%, followed by GitHub Copilot (67.9%), Google Gemini (47.4%), and Claude Code (40.8%), per Stack Overflow. GitHub Copilot separately reports over 20 million total users.

How many people use GitHub Copilot?

GitHub Copilot crossed 20 million users as of Microsoft’s Q4 FY25 earnings in mid-2025, up more than 5 million in one quarter. Over 90% of the Fortune 100 have adopted it, per GitHub’s leadership.

Why don’t developers trust AI coding tools?

The main reason is near-misses. 66% of developers cite “AI solutions that are almost right, but not quite” as their top frustration, and 45% say debugging AI code takes longer than writing it, per Stack Overflow. Overall, 46% now distrust AI accuracy.

Are AI coding tools better for junior or senior developers?

Junior developers gain more. GitHub’s study found the largest speedups among less-experienced developers, while METR found AI slowed experts working in codebases they already knew well. The tool front-loads knowledge a newcomer lacks.