Loading today’s stories
Build1 distinct publisher3 min readUpdated
JetBrains reports 90% of professional developers use coding agents weekly. The number that matters for tooling bets is not adoption but conversion, where Claude Code is far ahead.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
JetBrains has published agent adoption figures from its Developer Ecosystem Survey 2026, which it describes as a globally representative survey of more than 15,000 professional developers, now in its tenth year [1]. Between May and July 2026, 90 percent of those developers used AI coding agents at work at least weekly and 68 percent used them daily [2], which moves the procurement question past whether and onto which one, and who owns the bill.
The lead has changed hands. Claude Code was used at work by about 39 percent of professional developers worldwide in May-July 2026, up from 18 percent in January 2026 [3], while GitHub Copilot slid to 21 percent from 29 percent a year earlier [4]. That is a ratio of roughly 1.9 to 1, which JetBrains rounds to "twice as often" [1][5], and a relative decline of about 28 percent for the tool JetBrains credits with bringing AI-assisted coding into the spotlight in 2023 [3][6]. In the United States, Claude Code sits at 47 percent [7].
The figure worth carrying into a planning meeting is conversion, not headline adoption. JetBrains reports that 31 percent of developers name Claude Code as their single most-used AI coding tool, an almost 80 percent conversion rate from regular workplace use [8]. Run the same arithmetic on the others: Copilot has 79 percent awareness against 21 percent adoption, about 27 percent [9][4], and Codex has 65 percent awareness against 16 percent adoption, about 25 percent [10][5]. Awareness is not the bottleneck. Displacing whatever the developer already opens is.
Codex is the one moving fastest from behind: 3 percent adoption in January to 16 percent in May-July, with awareness up from 27 percent to 65 percent [10][11]. Cursor went the other way, from 18 percent to 12 percent adoption even as mind share rose from 69 to 75 percent [12], a relative drop of about a third [6], with the steepest fall in China, from 28 to 16 percent [13]. Google Antigravity held at 6 percent adoption while awareness climbed from 29 to 47 percent, and reached 15 percent in India, up from 10 percent [14][15]. OpenCode, with no large vendor attached, is at 7 percent adoption and 42 percent mind share [16].
Note that the reported shares add to about 110 points across seven tools [7], so multi-agent workstations are the norm rather than the exception, and any policy written as a single-vendor standard will be honoured selectively. JetBrains has an obvious interest here: it says 39 percent of Copilot users run it inside JetBrains IDEs, roughly 8 percent of all developers [17][8], puts JetBrains AI and Junie at about 9 percent [18], and is shipping Air, an agentic development environment in preview, plus JetBrains Central as a control plane, with Claude Agent, Codex, Copilot and OpenCode wired into IDE chat and others reachable via the Agent Client Protocol [19][20]. The survey and the multiplexer are sold by the same company.
Watch whether Claude Code's roughly 80 percent conversion rate holds through the next reading, because that, not adoption, is what makes a seat renew. Watch Codex conversion specifically: awareness has already been solved, so the next six months test the product. And watch Copilot's floor, since 86 to 90 percent awareness in Europe, the UK and the US [9] plus enterprise distribution can hold a tool in place long after it stops being anyone's first choice.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
JetBrains' Developer Ecosystem Survey 2026 is described as a large-scale, globally representative study of more than 15,000 professional developers worldwide, in its tenth year, run by JetBrains' Strategic Research and Market Intelligence team.
In May-July 2026, around 39% of professional developers worldwide were using Claude Code at work, up from 18% in January 2026.
GitHub Copilot adoption declined from 29% a year ago to 21% in May-July 2026.
In the United States, Claude Code adoption at work is 47%.
JetBrains is expanding its agentic ecosystem with Air, an agentic development environment in preview that combines multiple coding agents in one workflow, and JetBrains Central, a unified control and execution plane for agent-driven development.
As of May-July 2026, 90% of professional developers were using AI coding agents at work at least weekly (local or remote cloud agents), with 68% using them daily.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Large single-vendor survey, no error bars or replication
The underlying instrument is unusually large (15,000+ professional developers) with published quota, localisation and reweighting methodology, which is above the bar for most tooling-share claims. But every number in the cluster traces to one self-interested publisher, with no margins of error, no per-country sample sizes, no question wording, mixed comparison baselines (Copilot 'a year ago' versus January 2026 for others), and no independent tracker or vendor-reported telemetry to cross-check. The derived ratios are arithmetic on the same instrument, not corroboration.
Near-universal weekly agent use, multiple tools past double digits
This is a story about adoption and the disclosures are dense: 90% weekly and 68% daily agent use, with Claude Code at 39%, Copilot 21%, Codex 16%, Cursor 12%, JetBrains AI/Junie 9%, OpenCode 7% and Antigravity 6%, plus shipped IDE integrations for four agents and ACP attachment for dozens more. Shares summing to ~110 points show stacking rather than substitution. The reservation is that all figures are self-reported survey behaviour rather than measured deployments or billing data.
Directionally supported, mildly overstated by superlatives and single sourcing
The core comparison holds arithmetically: 39% versus 21% is about 1.9x, so 'twice as often' and the cluster's two-to-one headline are close to the data. The overstatement is in register and sourcing rather than in the numbers - 'unprecedented pace', 'by far the most widely adopted' and 'lost its leadership' are vendor superlatives resting on one unreplicated instrument with no error bars, and the report's market-structure narrative sits next to JetBrains' own products and preview roadmap. Claims about Air and Central carry no availability, pricing or usage evidence at all.
Vendor-authored market research adjacent to its own product line
JetBrains publishes the survey and sells in the market it measures: the same post reports JetBrains AI/Junie at 9%, highlights that 39% of Copilot users reach it inside JetBrains IDEs, promotes ACP-based neutrality, previews Air and Central, and links to JetBrains' AI offerings for teams and organisations. That gives a clear interest in a narrative where no single agent vendor owns the surface and the IDE is the aggregation layer. The methodology disclosure and reweighting for JetBrains-product familiarity are mitigating, but no conflict statement accompanies the market-share claims.
Trust the direction, discount the decimals
Confidence is limited by structure, not by sloppiness: the sample is large and the method partially documented, and the internal arithmetic checks out, so the direction of travel (Claude Code ahead, Copilot and Cursor receding on usage while retaining awareness, Codex climbing fast) is reasonably firm. But with one interested publisher, no precision statistics, inconsistent baselines and no external replication, exact levels and small gaps should not be treated as settled.
build
Waku 0.1.0 bets the product is the control plane, not another coding agent1 distinct publisher
product
Wu says Cognition is not for sale. The more useful fact is who bought Cursor last week.1 distinct publisher
security
The model is now choosing the extortion targets, not just writing the malware1 distinct publisher
product
Binance gives agents a trading seat, and gives users the permission slip1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 18, 2026