JetBrains Surveyed 10,000 Developers and Copilot's Dominance Quietly Cracked — Reading Claude Code's 6x Rise as a Once-Struggling Engineer

[[IMG:hero]]

I’m Gen. This week, a set of numbers came out that felt like a reckoning — the kind that makes you stop and check your assumptions.

JetBrains released the second installment of their “AI Pulse” survey in April 2026, based on data collected in January 2026. Over 10,000 professional developers answered the same set of questions, and the results were published on the JetBrains Research blog (JetBrains Research, 2026-04).

Three numbers I want to put on the table first.

• GitHub Copilot: 76% awareness, 29% active business use. It’s still the most recognized and most used AI coding tool.
• Claude Code: 57% awareness, 18% active business use. Usage grew roughly 6x in under a year.
• That said, only 18% of all developers surveyed are actively and regularly using any AI coding tool at work.

Some people will summarize this as “Copilot’s one-horse dominance has cracked.” My read is different. It hasn’t cracked — it’s hit a ceiling. And the reason that ceiling formed is that Claude Code has been pushing hard from below at a steep angle.

What follows is my attempt to make these four JetBrains data points three-dimensional, using my own hands-on experience and Accel VC’s coexistence thesis. At the end of the article, I want you to walk away with three concrete actions to update your own tool choices this week.

[[IMG:1]]

First, Let the Numbers Breathe — The 4 Lines from JetBrains That Actually Matter

Let me lay out the full picture from the JetBrains survey before I start filtering it. I’ll narrow things down later — just take in the overall shape for now.

Some context first. The sample is 10,000+ professional developers. The survey was conducted in January 2026 by JetBrains Research, under the name “AI Pulse,” second edition. The first edition ran in September 2025, with a baseline from April–June 2025. Because they asked the same questions at three points in time, you can actually see which direction things are moving.

The four key lines.

First: awareness. GitHub Copilot at 76%, ChatGPT (for coding) at 68%, Claude Code at 57%. Cursor also shows up near the top, but JetBrains didn’t publish Cursor’s specific percentage — just that it ranked close behind Claude Code (JetBrains Research, 2026-04).

Second: active business use. GitHub Copilot at 29%, Claude Code at 18%. The gap between “aware of it” and “actually using it at work” is 47 percentage points for Copilot and 39 for Claude Code. In other words, more than half of awareness is “heard of it but never tried it.”

Third: Claude Code’s growth curve. Awareness was 31% in April–June 2025, rose to 49% by September 2025, and reached 57% in January 2026. Active business use went from roughly 3% at the first data point, through 1.5x growth, to 18%. JetBrains’ report explicitly states “1.5x vs. September 2025, and 6x vs. April–June 2025.”

Fourth: Claude Code’s loyalty metrics. CSAT (customer satisfaction) at 91%, NPS (net promoter score) at 54. JetBrains describes these as “the highest product loyalty metrics in the market.” I’ll go into what this feels like from personal experience in a later section, but the simple truth behind this number is: people who use it don’t let go.

When you let the numbers breathe, here’s the picture that emerges. Copilot is “broad and shallow.” Claude Code is “deep and fast-growing.” This isn’t market polarization — it’s a new layer forming on top.

That’s the data dump. In the next section, I’ll insert my own judgment into the narrative of “Copilot’s dominance has cracked.”

[[IMG:2]]

“Copilot’s Dominance Has Cracked” — But the Reason Has Nothing to Do with Copilot

The phrase “Copilot’s dominance has cracked” is half right and half wrong.

The right half first. JetBrains’ own language is that Copilot’s growth “has stalled since last year” — both in awareness and adoption. The same company asked the same question at three points in time, and the upward curve went flat. That’s a data fact.

The wrong half: “stalled” doesn’t mean “cracked.” Copilot still leads both metrics at 76% awareness and 29% business use. What cracked was the assumption of a runaway win — not Copilot’s actual position.

This is where engineers like me — recovering from years of being overwhelmed by tools — tend to make a mistake. If you see “Copilot growth stalled, Claude Code surging,” you want to jump to “time to switch.” But that’s not what the data shows. Copilot users haven’t en masse migrated to Claude Code. The growth is coming from net new entrants — people who weren’t using AI coding tools before, coming in through Claude Code. That’s my read.

Two pieces of evidence.

One: the combined active use rate. Copilot 29% + Claude Code 18% = 47%. But JetBrains’ report also says “developers who actively use any AI coding tool at work: 18%.” There’s no contradiction here if you think carefully about the denominators. The 29% and 18% per-tool figures are measured against the base of those aware of each tool; the overall 18% is measured against all developers. JetBrains doesn’t make the methodology explicit, so I’ll note both interpretations.

Two: the shape of Claude Code’s growth. 1.5x in six months, 6x in under a year. You can’t get there by migrating existing Copilot users. The growth rate implies a large share of net new adopters.

The more useful question isn’t “should I switch from Copilot to Claude Code?” It’s “which layer of work am I mostly doing?” Keep that in mind. If you’re only on Copilot, the option is “try adding Claude Code.” If you’re not using AI coding tools at all, the question is which tool to start with. Don’t box yourself inside a cracked assumption.

CSAT 91%, NPS 54 — The Difference Between “Helps You Move” and “Stays Until It’s Fixed”

Let me translate Claude Code’s 91% CSAT and 54 NPS into something concrete. I said earlier that people who use it don’t let go. Here’s why — from seven months of daily use.

Before I switched, I was on GitHub Copilot for two years. The shift in experience becomes most obvious the moment something breaks.

Copilot’s core job is suggesting the next line or few lines. Its suggestion quality is high, and its IDE integration is genuinely well-built. But when an error hits and I need to fix it, I had to drive. Read the error log, form a hypothesis, write the correction. Copilot would assist that process, but it wouldn’t run it for me.

Claude Code was different. When an error comes up, I paste the terminal output and ask. It reads the relevant files, identifies the probable cause, writes a patch, runs the tests, and sees things through until they pass. My role shrinks to three commands: “please do this,” “go with this approach,” and “did the tests pass?”

If I had to say it in one sentence: Copilot is “a partner who helps you move.” Claude Code is “a partner who stays until the thing is actually fixed.”

A 91% CSAT reflects that experience gap. It’s not “the next line was accurate” — it’s “it stayed until the error was gone.” A partner who closes the distance to completion earns high marks.

But this isn’t a universal win. Copilot has real strengths. It’s always resident in your IDE, serving candidates without you having to think about launching anything. For work that lives entirely inside the editor, Copilot is faster than spinning up a Claude Code terminal session. In the “single-step completion” world, Copilot still feels stronger to me.

[[IMG:3]]

So Claude Code’s high ratings aren’t because it’s superior to Copilot — they’re because it’s doing a different job. IDE completion and agentic execution are different tools that should be used differently. The 91% CSAT is an evaluation of the agentic layer, not a verdict on inline completion.

”Only 18% of All Developers Use AI Coding Tools” — The 3 Barriers Keeping 82% Out

Now for the number in the JetBrains survey that I found most meaningful: “developers actively using any AI coding tool at work is still just 18%.”

76% awareness, but only 18% regular business use. The 58-point gap is stacked with people who know about these tools but aren’t using them.

As a former struggling engineer, I understand that 82% deeply. During the years I stepped away from code, I read dozens of “AI will transform development” articles. I read them. I didn’t try the tools. Why?

I think there are three reasons.

First: no clear mental picture of how it fits into your actual day job. “Copilot is useful” means nothing until I can see exactly where it slots into my workflow. AI coding tools get treated abstractly, and the translation from “cool technology” to “my specific work” takes time.

Second: company security policies and costs. Getting a tool approved at an organization involves internal sign-offs, data handling policies, and monthly billing — hurdles that exist before any individual can even try the thing. JetBrains’ report explicitly names “enterprise security policies” and “integration costs” as adoption barriers.

Third: distrust of AI-generated code. This one I feel in my bones. “It ran. But I don’t understand why it ran” is genuinely scary. The moment of success is less concerning than what comes after — debugging and maintenance on code you don’t fully comprehend.

The 82% aren’t lazy or uninformed. Three real barriers exist.

The key point is that these three barriers can’t be knocked down by showing people impressive numbers. Copilot at 76%, Claude Code at 57%, 6x growth — these prove the market is moving. But they don’t answer: how does it fit your specific job, can you get it past your IT team, and do you trust code you didn’t write? None of those questions get answered by a chart.

That’s why the next section brings in Accel VC’s coexistence argument. I want to give you a decision framework for your own situation, not push you to act on market momentum.

Accel VC: “The Market Is Big Enough for Both” — Compressing the Numbers With the Coexistence Thesis

Just before the JetBrains survey dropped, another important statement came out.

Miles Clements, a partner at venture firm Accel (an early investor in Cursor), appeared on the “20VC” podcast in March 2026. His central claim was blunt: “The idea that Cursor is dead is completely false.” He pushed back hard on the binary framing that’s been applied to this market (AOL News, 2026-03 (via Business Insider), 20VC official episode page).

Clements said two things that matter in this context.

“This market is growing enormously, and I don’t think a lot of these companies are actually experiencing success at the expense of the others.”

(On Claude Code:) “Amazing product.”

A VC who invested in Cursor going out of his way to call Claude Code “amazing” is remarkable given his financial position. My interpretation: the market growth rate is high enough that the worry is seat-taking, not mutual destruction.

Back to the JetBrains numbers. Combined active use: roughly 47% (Copilot 29% + Claude Code 18%). But overall developer market penetration for any AI coding tool: 18%. The gap represents either multi-tool usage, denominator differences, or both. Either way, the market has significant room to grow.

If you frame it as “Copilot vs. Claude Code,” one winning means the other losing. If you frame it as “single-step completion vs. agentic execution,” both can grow simultaneously.

[[IMG:4]]

The most important thing I want to say today: don’t let market numbers push you into a “should I switch” panic. Instead, measure what fraction of your coding time is spent on single-step completion versus agentic execution work.

For me, in my tooling development work: roughly 30% single-step completion, 70% agentic execution. That’s why my Claude Code dependency is high. But for someone whose work is primarily frontend UI polish, those ratios might be reversed.

The decision axis isn’t “market share.” It’s “the nature of your specific work.” That’s the conclusion I get when I overlay the Accel VC statement on top of the JetBrains survey data.

This Week’s Move — Translating the Survey Into Your Own Tool Decision

I’ve put enough numbers on the table. Let me compress them into actions you can start this week.

Three actions. All designed to be startable within seven days.

Action 1: Track your “coding time allocation” for one week

This week, roughly sort your coding time into three buckets:

• Single-step completion time: work that resolves in a few lines, entirely within the IDE
• Agentic execution time: multi-file edits, error resolution, test additions
• AI-free design time: spec review, requirements, whiteboarding

Paper, notes app, doesn’t matter. One week of data will show you your actual allocation.

Action 2: If you’re not on Claude Code yet, try it on one task

Use the “18% active use” statistic as a nudge — not a mandate — to try Claude Code on one task. Good candidates: “a refactor that touches multiple files” or “diagnose the root cause of a recent error log.” Trying it in a territory where Copilot’s inline completion feels limited will give you a direct feel for what the 91% CSAT is about.

One important prerequisite: check your company’s security policy first. Whether the tool sends code to external servers, how logs are handled — these are checklist items before any tool selection decision.

Action 3: Explicitly consider the option of not switching

If Copilot is working for you, there’s no urgent reason to change. If Action 1 shows your allocation is 80% single-step completion, staying on Copilot is the rational choice. Markets can move without that being a mandate for you to move with them.

The JetBrains survey and the Accel VC statement never say “switch.” They say “the market is growing” and “there’s room for both.” I’m taking home the same discipline they model: read the numbers without being swept by them.

Summary

JetBrains’ survey of 10,000 developers was one of the most valuable market checkpoints I’ve seen.

Key points:

• GitHub Copilot: 76% awareness, 29% active use. Still on top — but growth has stalled
• Claude Code: 57% awareness, 18% active use. 6x growth in under a year, 91% CSAT, NPS 54
• Only 18% of all developers actively use any AI coding tool at work. 82% face three real barriers: no clear fit, security/cost hurdles, distrust of AI-generated code
• Accel VC coexistence thesis: single-step completion and agentic execution are different markets; both have room to grow
• Decision axis: not “market share” but “the nature of your own work”

The feeling of “a master engineer living inside me” — I got that back through Claude Code. Given that I’m a former-struggling-engineer saying it, take it with a grain of salt. But the 91% CSAT isn’t an accident.

Even so, this article never said “switch.” If Copilot works for you, keep using it. If you’re not on AI coding tools at all, you don’t need to rush. The 82% barriers are real, and rushing past them isn’t the answer.

Read the numbers to avoid being swept by the numbers. That’s the healthiest way to digest this survey.

This week, when you have your “coding time allocation” memo ready — share it with me. My DMs are open, and comments work too. Once I see the allocation, I’ll help you figure out the next move.