JetBrains published their developer tools research this week. The headline number: 90% of developers regularly use at least one AI tool at work. That's an extraordinary adoption rate for any technology, let alone one that barely existed three years ago.
But the more interesting finding was buried further down: Cursor and Claude Code now share second place in workplace adoption at 18% each. GitHub Copilot still leads. OpenAI Codex is gaining. xAI just entered the race with Grok Build.
Here's the thing nobody is saying clearly: the developers getting the most out of AI aren't picking the best tool. They're running a deliberate stack of two or three tools, each doing what it's actually good at. And if you're still treating this as a one-tool decision, you're leaving significant productivity on the table.
Why One Tool Isn't Enough
Every major AI coding tool right now has a different strength. Cursor is exceptional at understanding large codebases and making targeted edits across multiple files. Claude Code is strong at multi-step reasoning, ambiguous tasks, and working through problems that require judgment rather than just code generation. Copilot is fast, integrated, and good at autocomplete-style assistance in the flow of writing.
These aren't redundant. They're complementary. Using only one of them because you're trying to simplify your setup is like using only a hammer because you don't want to carry a toolkit. You'll get the job done more slowly, with worse results.
The Stack That's Actually Working in 2026
Based on how the community has been talking about this — and what the JetBrains data confirms — there's a pattern emerging. It looks something like this:
- Claude Code (or a comparable reasoning-heavy agent) for planning, architecture decisions, and multi-step tasks where you need to think through the problem before touching the codebase.
- Cursor for active development — navigating a large repo, making cross-file edits, understanding how existing code connects together.
- Copilot or inline autocomplete for the flow state work — filling in function bodies, suggesting variable names, completing patterns. Low friction, always on.
The key insight
Each tool covers a different cognitive mode: reasoning and planning, navigating and editing, and completing and flowing. When you map tools to modes instead of trying to find one tool that does everything, your workflow gets dramatically faster.
The Orchestration Problem
The challenge with running a stack is knowing when to switch. The temptation is to stay in one tool because context switching has a cost — you have to re-explain the problem, re-share the relevant code, re-establish what you were trying to do.
The developers who've figured this out have solved the context switching problem rather than avoiding it. They maintain a short working document — sometimes just a few bullet points in a scratch file — that captures the current task, the relevant constraints, and the decisions already made. This becomes the handoff context when they switch tools.
It sounds like overhead. In practice it takes 30 seconds to write and saves five minutes of re-explaining every time you switch. It also makes you think more clearly about what you're actually trying to do, which is valuable regardless of which tool you're using.
What About the New Entrants?
Google launched Managed Agents at I/O this week. xAI launched Grok Build. Alibaba released Qwen3.7-Max specifically targeting agentic coding. The tool landscape is expanding, not consolidating.
The instinct is to wait and see which one wins. That's the wrong frame. None of them will win outright — they'll each find the part of the stack they're best at and settle there. The question to ask about any new entrant isn't "is this better than what I have?" It's "does this do something meaningfully better than what I already have for a specific type of task?"
The 10% Who Aren't Using AI Yet
The JetBrains number that nobody's talking about is the other side: 10% of developers still don't regularly use AI tools at work. Some of that is by choice. Some of it is organisational. Some of it is just not having found the right entry point.
If you're in that 10%, the stack framing is actually a good way in. Don't try to adopt everything at once. Pick one tool for one specific type of task — code review, writing tests, explaining unfamiliar code — and use it deliberately for two weeks. Then add the next layer. The developers seeing the biggest gains didn't change their whole workflow overnight. They added one thing, made it habitual, then added the next.
“The best AI workflow isn't the one with the best single tool. It's the one you'll actually use consistently.”
Start With the Seams
If you're looking for where to start building a proper AI coding stack, look for the seams in your current workflow — the places where you slow down, get stuck, or context-switch manually. Those are the places where an AI tool adds the most value with the least disruption to everything else that's working.
The tool landscape will keep changing. New entrants, new benchmarks, new capabilities every month. The developers who will keep winning aren't the ones who pick the best tool today. They're the ones who've built the habit of using tools deliberately and adding new ones when they solve a real problem in their workflow.