Vibe coding means describing what you want in plain language and letting an AI write the code. AI coding assistants are the wider category: autocomplete, chat and agents that help developers write, review and fix code.
Almost every developer survey now shows heavy use of these tools. The same surveys show a gap between how much people use them and how much they trust them.
Vibe coding and AI coding statistics
| Metric | Figure | Source and date |
|---|---|---|
| Software professionals using AI | 90% (up 14% on the prior year) | Google DORA report, Sep 23, 2025 |
| Developers regularly using AI tools | 85% | JetBrains survey (24,534 respondents), fielded Apr to Jun 2025 |
| Developers using AI tools daily | 47.1% of all respondents; 50.6% of professionals | Stack Overflow 2025 survey |
| AI coding tool users who use them daily | 73% | Stack Overflow 2026 survey, Oct 6, 2026 |
| Developers who say they vibe code | About 15% in some form; 72.2% say no | Stack Overflow 2025 survey |
| Trust in AI output accuracy | 32.7% trust; 45.7% distrust | Stack Overflow 2025 survey |
| New code at Google generated by AI | About 75% | Reported Apr 22, 2026 (TechSpot) |
| GitHub Copilot all-time users | 20 million+ | Microsoft earnings call, Jul 30, 2025 |
| Insecure code in AI output | 45% of test cases | Veracode, Jul 30, 2025 |
| Task time with AI vs without (experienced devs) | 19% longer | METR, Jul 10, 2025 |
Where “vibe coding” came from
- Andrej Karpathy posted the definition on X on February 2, 2025. He described a style where he works with Cursor Composer by voice and “barely even touches the keyboard” (Know Your Meme, which quotes the post).
- Collins Dictionary named “vibe coding” its Word of the Year for 2025, announced November 6, 2025. Collins defines it as an emerging form of software development in which AI turns plain-language instructions into code (News On AIR report).
How many developers use AI coding assistants?
Three large surveys land in a similar place, though each measures something slightly different.
- DORA 2025 (Google): About 5,000 technology professionals were surveyed. 90% use AI at work, up 14% from the year before, and the median is two hours a day. About 65% say they rely on it heavily (Google, Sep 23, 2025).
- JetBrains 2025: 85% of developers regularly use AI tools for coding and development, out of 24,534 respondents in 194 countries. About 68% expect their employers to require AI tool skills soon (JetBrains Research, survey run April to June 2025).
- Stack Overflow 2025: 47.1% use AI tools daily, and 30.9% use AI agents at work. Another 17.4% plan to adopt agents, and 37.9% have no plans to (Stack Overflow, Jul 2025).
- Stack Overflow 2026: Among respondents who use AI coding assistants or agents, 73% use them daily. Use is spread across time: 31% use AI four or more hours a day, 24% use it two to four hours, and 26% use it one to two hours. The survey had 30,000+ respondents and was published October 6, 2026.
These figures are not interchangeable. “Use AI at work” (DORA), “regularly use” (JetBrains) and “use daily” (Stack Overflow) are different bars.
Stack Overflow’s 2026 page, as I could read it, did not state an overall share of developers using AI, so I have not compared it with 2025.
Vibe coding vs professional practice
The most useful stat for anyone writing about vibe coding is how few professionals say they do it.
- 72.2% of respondents say vibe coding is not part of their professional work, and 5.3% say so emphatically. About 15.1% say yes in some form (Stack Overflow 2025 survey).
For the platforms built around the idea, see the per-platform pages for Lovable and Replit.
AI coding assistant market and usage by tool
- GitHub Copilot: Satya Nadella said Copilot passed 20 million all-time users on the July 30, 2025 earnings call. A GitHub spokesperson confirmed to TechCrunch that this is an all-time total, not active users. Microsoft also said 90% of the Fortune 100 use it. No paid-subscriber number was given.
- GitHub overall: 180 million+ developers on GitHub, with 36 million+ joining in a year (up 23%), and nearly 80% of new developers using Copilot in their first week (GitHub Octoverse, Oct 28, 2025). The Copilot coding agent created 1 million+ pull requests between May and September 2025.
- Cursor: An AI code editor that SpaceX agreed to buy for $60 billion in stock. One news aggregator reports the deal closed on August 14, 2026; I found no second source, so confirm before quoting it. Details are in our Cursor statistics article.
- Claude Code: An AI news aggregator, citing CNBC, reported it reached $1 billion in annualized revenue within six months of launch (AI Weekly, May 20, 2026). Anthropic’s own run rate was reported at more than $30 billion in Q1 2026. Treat both as secondhand until you see CNBC’s original.
How much code is written by AI
- Google: About 75% of new code is generated by AI and reviewed by engineers, as announced April 22, 2026. In October 2024 the company said about a quarter. By late 2025 it had said about 50% (TechSpot, Apr 2026). TechSpot did not name who made the announcement, so cite it as “Google, as reported by TechSpot”. Definitions differ. “Generated by AI” is not the same as “shipped without edits”, and each company counts its own way.
Trust and frustration
Usage is high, and confidence is lower.
| Finding | Figure | Source and date |
|---|---|---|
| Trust AI output accuracy | 32.7% (3.1% highly) | Stack Overflow 2025 |
| Distrust AI output accuracy | 45.7% (19.6% highly) | Stack Overflow 2025 |
| Top frustration: “almost right, but not quite” | 66% | Stack Overflow 2025 |
| Debugging AI code takes more time | 45.2% | Stack Overflow 2025 |
| Great deal or a lot of trust in AI | 24% | Google DORA, Sep 2025 |
| Trust a little or not at all | 30% | Google DORA, Sep 2025 |
| Need source attribution to trust AI output | 93% | Stack Overflow 2026 |
| View AI at least somewhat positively | 62% (71% of daily users) | Stack Overflow 2026 |
JetBrains lists the top concerns as inconsistent quality of AI code, limited understanding of complex logic, privacy and security risks, harm to their own coding skills, and weak context awareness.
Productivity: what the research says
The evidence points in two directions.
- Self-reported gains: Over 80% of DORA respondents say AI improved their productivity, and 59% report a positive effect on code quality. DORA also found AI adoption linked to higher delivery throughput, reversing its previous year’s finding (Google, Sep 2025).
- Measured slowdown in one study: METR ran a randomized trial with 16 experienced open-source developers on 246 real tasks. With AI tools they took 19% longer. They had predicted a 24% speedup and still believed they were about 20% faster afterward (METR, Jul 10, 2025, reported by The Decoder, Jul 11, 2025).
The METR sample is small, the tasks were in mature codebases the developers knew well, and the tools were early-2025 models. It does not prove AI slows every developer. It shows that feeling faster and being faster can differ.
Code quality and security
- Veracode tested 100+ LLMs on 80 coding tasks and found AI introduced security vulnerabilities in 45% of cases. Java had the highest risk at over 70%, while Python, C# and JavaScript ranged from 38% to 45% (Veracode, Jul 30, 2025).
- CodeRabbit analyzed 470 open-source pull requests and found AI-generated ones averaged about 10.83 issues versus 6.45 for human-written ones, roughly 1.7 times more. AI pull requests had about 1.4 times more critical issues and 1.57 times more security findings. Specific flaws included cross-site scripting (2.74 times), insecure object references (1.91 times) and improper password handling (1.88 times). Humans had more spelling errors (1.76 times). CodeRabbit sells code review tools, and it notes it cannot be sure the “human” pull requests were entirely human-written (CodeRabbit, reported by The Register, Dec 17, 2025).
Both reports are about code assistants in general, not every vibe-coded app.
Both are a reason to review auth, payments and data handling before launch. If a prototype is going to production, a developer review or custom web application development is the safer route.
Forecasts
- Gartner expects enterprises to use vibe coding techniques and tools to create 40% of new production software by 2028, as reported by CIO Dive on June 11, 2025. CIO Dive did not link the original Gartner research, so confirm the wording against Gartner before using it.
- Gartner forecast worldwide AI spending of $2.52 trillion in 2026, up 44%, with AI software rising from $283.1 billion to $452.5 billion (Gartner, Jan 15, 2026). This is overall AI spending, not a measure of coding tools alone.
I did not find a verified market-size figure for AI coding tools specifically. Several blogs quote one without a primary source, so I left it out.
What the numbers suggest?
- Usage is mainstream, vibe coding is not. Most developers use AI, and fewer than one in six say they vibe code.
- Trust trails usage. Heavy daily use sits beside high distrust, and “almost right” is the most common complaint.
- Large companies report the highest AI code shares. Google’s 75% figure and Copilot’s Fortune 100 reach (both company statements) point that way.
- Review remains the bottleneck. Both security studies point to checking output instead of shipping it as is.
FAQ
What percentage of developers use AI coding assistants?
Between 85% and 90% in the large 2025 surveys by JetBrains and Google DORA, depending on how “use” is defined. Stack Overflow’s 2025 survey found 47.1% use AI tools daily.
How many developers vibe code?
In Stack Overflow’s 2025 survey, about 15.1% said yes in some form and 72.2% said no.
How much code does AI write?
Google said about 75% of its new code is AI-generated and reviewed by engineers (reported April 2026). Other companies count differently.
How many people use GitHub Copilot?
Microsoft reported 20 million+ all-time users on July 30, 2025. That is not a count of active users.
Is AI-generated code secure?
Not by default. Veracode found security flaws in 45% of test cases, and CodeRabbit found 1.57 times more security findings in AI pull requests.
Do AI coding tools make developers faster?
Most developers say yes. METR’s 2025 trial found experienced developers took 19% longer, so the evidence is mixed.
Sources
- Google: Inside our 2025 DORA report (Sep 23, 2025)
- JetBrains Research: State of Developer Ecosystem 2025 (survey Apr to Jun 2025)
- Stack Overflow 2025 Developer Survey: AI (Jul 2025)
- Stack Overflow: Results of the 2026 Developer Survey (Oct 6, 2026)
- TechSpot: Google AI now generates 75% of new code (Apr 2026)
- TechCrunch: GitHub Copilot crosses 20 million all-time users (Jul 30, 2025)
- GitHub Octoverse 2025 (Oct 28, 2025, updated Feb 28, 2026)
- AI Weekly: Anthropic hits $30B run rate (May 20, 2026, citing CNBC)
- AI Weekly: SpaceX closes $60B Cursor deal (Aug 15, 2026)
- Veracode press release (Jul 30, 2025)
- The Register: AI-authored code contains worse bugs (CodeRabbit) (Dec 17, 2025)
- The Decoder: METR study (Jul 11, 2025)
- CIO Dive: vibe coding and enterprise strategy (Gartner forecast) (Jun 11, 2025)
- Gartner: Worldwide AI Spending Will Total $2.5 Trillion in 2026 (Jan 15, 2026)
- Know Your Meme: Vibe Coding (Karpathy post, Feb 2, 2025)
- News On AIR: Collins Word of the Year 2025 (Nov 6, 2025)
