Half of developers now let AI write their app logic: what JetBrains' 2026 survey found

AINews

Half of developers now let AI write their app logic: what JetBrains' 2026 survey found

Task board on a dark brick wall, with columns of coloured sticky notes
Photo by cottonbro studio on Pexels

JetBrains' latest survey data shows developers delegating far more to AI agents than a year ago, but most still hold onto debugging and code review. The numbers, what they mean, and what to take from them as a builder.

In this article
  1. What developers will hand to AI
  2. How much code agents actually write
  3. How developers feel about it
  4. What to take from it as a builder
  5. About the data

"100% of my code is written by AI" is a common brag on X. JetBrains decided to measure what's actually happening, and the numbers show a big shift in one year. The data comes from its Developer Ecosystem and AI Pulse surveys, published by JetBrains Research this month and in August.

What developers will hand to AI

Share of developers ready to delegate each task to AI:

Task Apr–Jun 2025 May–Jul 2026
Writing application logic 18% 50%
Writing tests 50% 75%
Refactoring 38% 60%
Code review (fully) 33% 43%

Debugging is the clearest holdout: 54% are not ready to delegate it, though that's down from 76% a year ago. JetBrains expects this to shift "as agentic debugging improves".

How much code agents actually write

In May–July 2026, JetBrains asked 15,000+ professional developers what share of last month's work code was agent-written, AI-assisted or fully manual. On average:

  • ~47% fully written by agents
  • ~38% written with some AI assistance
  • ~27% written fully by hand

(Those averages add up to more than 100% because people answered in ranges.)

Other findings:

  • Over half write less than 20% of their code manually, and one in five writes none without AI help.
  • Only about 22% rely almost entirely on agents (over 80% agent-generated).
  • Seniors lead: about a quarter of senior developers generate over 80% of their code with agents, more than juniors.
  • By tool: 32% of developers whose main tool is Claude Code generate over 80% of their code with agents, versus 42% for Codex users. JetBrains puts Claude Code's adoption at work at 39%.
  • By language: Go, JavaScript and TypeScript are the most agentic (54–55% agent-generated on average). C and C++ are the least.
  • By region: 32–35% of developers in China, Japan and South Korea generate over 80% with agents, versus about 16% in Europe and the UK.

How developers feel about it

  • 79% believe employers will expect AI-tool proficiency, up from 59% in mid-2024.
  • 37% think AI coding will reduce demand for professional developers, up from 28% in 2024.
  • Nearly a quarter think AI will eventually replace human developers entirely, up from 8% in 2024. Among heavy agent users the figure is 37%.

What to take from it as a builder

  1. The gap is review. Developers happily hand over writing and tests but not reviewing or debugging. JetBrains itself recently wrote that "the bottleneck is no longer code generation; it's review." Get good at it with our guide to reviewing AI-written code and the code review skill.
  2. Tests are the easiest win. Three-quarters of developers already delegate them. If your agent isn't writing tests, start there.
  3. Context makes delegation work. Handing more work to an agent goes better when it knows your project's rules. An AGENTS.md is the cheapest upgrade.
  4. Pick a tool, then learn it deeply. See Cursor vs Claude Code vs Codex and our Codex CLI guide.

About the data

JetBrains says the survey is localized into eight languages, uses regional quotas, and is reweighted to represent the global developer population. About 90% of respondents are developers, programmers or software engineers. The tool-level comparisons come from respondents' own reports of their most-used tool. They describe who uses each tool, not how good the tools are.

FAQ

What share of developers let AI write code in 2026?

Per JetBrains Research, 50% of developers are ready to delegate writing application logic to AI, up from 18% a year earlier. In its 2026 survey, developers reported that about 47% of their code last month was fully agent-written on average.

What tasks do developers still not trust AI with?

Code review and debugging. Only 43% are ready to fully delegate code review, and 54% are not yet ready to delegate debugging, down from 76% holding out a year earlier.

Do developers think AI will replace them?

37% believe AI coding will decrease demand for professional developers (28% in 2024). Nearly a quarter believe AI will eventually fully replace human developers, up from 8% in 2024, and the view is more common among heavy agent users.

Which programming languages see the most AI-written code?

Per JetBrains, Go, JavaScript and TypeScript developers report the highest shares of agent-generated code (54–55% on average). C and C++ developers report the lowest.

Sources

  1. JetBrains Research, “Dynamics of Sentiment and Attitudes Toward AI Among Developers” (Oct 2026)https://blog.jetbrains.com/research/2026/10/dynamics-of-sentiment-and-attitudes-toward-ai-among-developers/
  2. JetBrains Research, “How Much Code Do Developers Really Let Agents Write?” (Aug 2026)https://blog.jetbrains.com/research/2026/08/how-much-code-do-developers-really-let-agents-write/

Newsletter

The Codema publication. Consent is required. You are not subscribed until you confirm from the email, and nothing is stored if email delivery is not connected.