AI Coding Tools Shift: Copilot Drops, Codex Soars, Claude Code Shines

The JetBrains AI Coding Agent Adoption Trends report landed this month, and it is the first survey in two years where the leaderboard actually moved. 15,000+ professional developers, surveyed May through July 2026, and three shifts land hard: Copilot lost 8 points of adoption, Codex went from 3% to 16% in six months, and Claude Code has the highest satisfaction score in the survey at 91% paired with the widest awareness-to-adoption gap in the market.

That is not noise. That is a market in motion, and it tells a specific story about what developers pay for once the novelty wears off. Here is what the data says, what it means for the tools you are choosing right now, and the part that does not move when the leaderboard does.

Copilot is sliding, and the receipt is in the numbers

GitHub Copilot dropped from 29% adoption to 21% in 12 months in the JetBrains data. Stack Overflow's separate survey shows the same direction, falling from 67% to 51% market share. Developer satisfaction collapsed from roughly 30% "most loved" to 9%.

Three concrete causes, in order of weight.

Billing is the loudest complaint. Some Pro+ subscribers are hitting their monthly usage cap by day two. At least one team reported a single file review consumed 20% of a monthly allowance. The free tier caps at 2,000 completions per month; Google Gemini Code Assist's free tier offers 180,000. That is a 90× gap on the entry tier, and it is the first thing a budget-conscious team lead sees when comparing.

Quality regressions since Q4 2025. The community has been documenting this independently for months — completions that over-trust file context, suggestions that used to nail patterns now miss. Copilot still has 79% mind share, so most teams have a license. But the gap between "we have it" and "we use it daily" is widening.

Competitive pressure at the free tier. When a serious alternative gives you two orders of magnitude more headroom for $0, the decision stops being about model quality. It becomes about who runs out first.

The honest read: Copilot is still the default at companies that bought it before the alternatives existed. New teams are picking differently.

Codex went from 3% to 16% in six months

That is not hype, that is pull. Codex did not grow because of a marketing campaign. It grew because developers actually switched.

The pattern is plain. Better raw model performance on the kinds of multi-file refactors developers run in their own repos. Integration with the IDEs most teams already ship. Pricing that does not punish a heavy day.

When 3% jumps to 16% in two survey cycles, that is not a sampling fluke — that is a workflow change. Developers are not just trying Codex. They are making it default for at least some classes of work.

The meaningful question is not "is it as good as Copilot?" It is "is it better than my current default on the work I actually do?" That is a benchmark you can run yourself in an afternoon.

Claude Code: 91% satisfaction, and the gap that explains it

Claude Code has the highest customer satisfaction score in the survey (91%) and the lowest awareness-to-adoption conversion. Awareness grew from 31% to 57% in a year — the fastest of any tool in the survey — and daily adoption is much lower than that number suggests.

JetBrains attributes this directly to the terminal-first interaction model. The data backs that up. In the US, where CLI-native workflows are more common, Claude Code adoption sits at 47%. Globally it is 18%. A 29-point split on the same product is a UX story, not a capability story.

The retention number is the part worth paying attention to. Among developers who become regular Claude Code users, 80% make it their primary tool. The funnel is narrow, but the people who get through it do not leave.

If you bounced off Claude Code because the terminal-first model felt like friction, that is a real reaction — not a sign the tool is bad. It is a sign it is not built for how you already work.

What the pattern actually says

Three tools, three different stories, but they converge on the same lesson. In 2026, developer loyalty tracks three things, in this order:

1. Will it run out before I am done? Billing caps and free-tier ceilings are now the first filter.
2. Can I get into it without re-learning my workflow? UX friction is the second filter, and it is why Claude Code's 91% CSAT does not translate to adoption.
3. Does it actually nail the work? Capability still matters — that is why Codex jumped — but it ranks below price and ergonomics.

That is a reordering from 2024, when novelty and raw model quality dominated the decision. The market is maturing.

How to actually compare them in your stack today

You do not need a four-week bake-off. You need a 90-minute swap test.

# 1. Pin a real task — a bug fix, a refactor, a doc rewrite — and run it
#    three times. Same prompt, same file, three tools. Save the diffs.

# 2. Track your current default's usage from its billing dashboard.
#    Note when caps hit.

# 3. Swap in the next tool via its native entry point (CLI, IDE plugin,
#    or API base URL). Use the same prompt. Same diff.

# 4. Score each on three axes: did it finish, how many tokens, did you
#    trust the diff enough to commit.

The decision is downstream of those numbers. If Copilot hits a cap before the third task, that is your free-tier data point. If Codex finishes twice as fast on the refactor, that is your migration case. If Claude Code feels like friction in your IDE-centric workflow, that is a real signal — not a personal failing.

The part that does not move when the leaderboard does

Here is the thing about a market this volatile. The AI tool writing your code is the most replaceable layer in your stack. Last year it was Copilot; this year Codex is climbing; next year it will be something else. The churn rate at the tool layer is now faster than the churn rate at the framework layer.

What does not churn is what those tools are writing code into. The component architecture, the cross-platform primitives, the place where generated code meets shipped UI — that is the durable substrate. If your templates are structured well, an AI agent can fill them in correctly today and you can swap the agent tomorrow without rewriting the app. If your templates are loose, every model swap is a rewrite.

Pick the tool that fits today's work. Build on the layer that survives tomorrow's swap.