When we first published this article in March, AI coding assistants were already common. Seven months later, the picture has shifted again: AI no longer just suggests the next line — agents now take on whole tasks, open pull requests and review each other's code. We have rewritten this guide with the latest 2026 data.
How many developers use AI now?
Nearly all of them. Two large 2026 surveys agree:
- Stack Overflow 2026 Developer Survey (30,000+ respondents): coding assistants and agents are the top AI use case at 66%, and 73% of people who use them do so daily (Stack Overflow).
- JetBrains Developer Ecosystem research (15,000+ professionals): 90% use AI coding agents at work at least weekly and 68% daily (JetBrains).
Which tools developers use
The tool landscape changed fast. Among developers who use coding agents, Stack Overflow found Claude Code used by 66% and GitHub Copilot by 59%. JetBrains' workplace data shows Claude Code's share more than doubling during 2026, with OpenAI's Codex also growing quickly:
| Tool (used at work) | January 2026 | Mid-2026 |
|---|---|---|
| Claude Code | 18% | 39% |
| GitHub Copilot | — | 21% |
| OpenAI Codex | 3% | 16% |
| Cursor | 18% | 12% |
Source: JetBrains Research, 2026.
How much code does AI write?
More than most people expect. JetBrains estimates that on average close to half of developers' work code is now fully generated by agents, and about 22% of developers rely on agents for more than 80% of their code (JetBrains). At the largest tech companies the share is higher: Google CEO Sundar Pichai said in April that around three-quarters of new code at Google is AI-generated and approved by engineers.
The key word is approved. AI writes the first draft; engineers decide what ships.
Does AI actually make developers faster?
This is the most debated question in the field, and the honest answer is "it depends on how you measure — and how you work".
- METR's 2025 randomised trial famously found experienced open-source developers were 19% slower with early-2025 AI tools, while believing they were faster. Its 2026 follow-up with newer agents pointed towards speed-ups, but METR itself says the data is unreliable because many developers refused to work without AI — so it is redesigning the study.
- Google Cloud's DORA research argues that returns on AI depend more on platform quality and team workflows than on the tools themselves, and often follow a "J-curve": an early dip, then gains (InfoQ on DORA).
Our experience matches the research: AI speeds up well-defined work — boilerplate, tests, migrations, debugging known issues — dramatically. It helps less on ambiguous design problems, and it can slow teams down if review and testing are weak.
Quality and security: the catch
AI writes code that looks right more often than code that is right.
- Veracode's GenAI code security research found that 45% of AI-generated code samples contained OWASP Top 10 vulnerabilities across more than 100 models.
- Developers know this: only 16% of daily AI users in the Stack Overflow survey trust AI across most tasks, and 93% say source attribution is needed to trust AI output.
- Few teams let AI near production: just 20% use it to deploy, operate or troubleshoot live systems.
Teams that benefit most invest in automated testing, AI-assisted code review, security scanning and strict review rules for AI-generated changes.
What changes for developers' jobs
AI is reshaping the role rather than removing it — but the effects are uneven:
- US software development job postings have rebounded since early 2025 but remain about 27.5% below pre-pandemic levels, and 71% of the recent increase came from senior roles (Indeed Hiring Lab).
- Stanford researchers found employment for developers aged 22–25 has fallen close to 20% from its 2022 peak, concentrated where AI automates rather than augments tasks (Stanford AI Index 2026).
- Skills in demand are shifting towards building and running AI systems — agents, RAG, MCP and evaluation — on top of solid engineering fundamentals. We break this down in what AI skills to learn in 2026.
The developer's job is becoming more about specifying, reviewing and integrating — writing clear requirements, judging trade-offs, and owning quality — and less about typing every line.
How AI fits into each stage of development
| Stage | Where AI helps today | What still needs humans |
|---|---|---|
| Requirements | Drafting user stories, spotting gaps | Deciding what to build and why |
| Design | Exploring options, generating diagrams | Architecture trade-offs, security model |
| Coding | Implementation, refactoring, migrations | Reviewing and owning the result |
| Testing | Generating unit and integration tests | Deciding what "correct" means |
| Code review | First-pass review, style and bug detection | Final approval, design feedback |
| Operations | Log analysis, incident summaries | Production changes and judgement calls |
How we use AI at Zinetics
We use AI coding agents across our projects, with a few rules that consistently pay off:
- Every AI change goes through human review and automated tests — no exceptions.
- Agents work in isolated branches with least-privilege access to secrets and infrastructure.
- We write clear specs first. Better instructions produce better code; vague prompts produce plausible bugs.
- We measure outcomes, not lines of code — cycle time, defects and rework.
The result for clients is faster delivery on the routine parts of a project, and more senior attention on architecture, security and the product decisions that matter.
What business leaders should do
- Expect faster delivery on well-defined work, but budget for review, testing and security.
- Ask your development partner how they use AI — and how they control quality and protect your code and data.
- Look beyond coding. The bigger opportunity is often putting AI into your product: assistants, automation and smarter search. See what's new in AI in 2026 for the wider landscape.
Zinetics combines AI-accelerated engineering with senior oversight to ship web, mobile and AI products faster without cutting corners. Let's talk about your project.