Key Takeaways
- Claude Code reached 39% adoption worldwide — nearly double GitHub Copilot, the former default, which fell from 29% to 21%.
- The fastest-growing coding tools all run in a terminal, not an IDE, and the stickiest one converts 80% of regular users.
- With agents increasingly driving deploys, hosts are retrofitting deploy paths, build logs, and guardrails to be readable by machines.
Nine in 10 professional developers now use an AI coding agent at work at least once per week, and 68% do so every day, according to results of the AI agent study published by JetBrains on August 18 based on its 2026 Developer Ecosystem Survey.
The survey is in its 10th year of operation and brings in data from more than 15,000 professional developers surveyed during the period between May and July. It represents the closest thing to a census for this industry — and may have simultaneously eliminated any doubts about the reality of AI-assisted coding.
As a result, the debate is over. The real news comes one level down: Which tool won? Because the ranking was reversed, not merely reordered.
Claude Code achieved adoption rates of 39% worldwide and 47% in the U.S., compared with 18% in January. That more than doubled in six months.
AI coding agent adoption at work
May–July 2026
“GitHub Copilot has lost its leadership, experiencing a decline from 29% adoption a year ago to 21% in May–July 2026,” the report says. This loss of market leadership was even more striking for a tool with access to the distribution capabilities of GitHub, the financial resources of Microsoft, and a head start of several years.
The degree of market awareness also was substantially greater for GitHub Copilot (79% worldwide and 86-90% in Europe, the U.K., and the U.S.) than for Cursor (75% awareness, with adoption sliding from 18% in January to just 12% at midsummer).
Adoption rates for Codex increased from 3% to 16% for the six-month period and awareness increased from 27% to 65%. Meanwhile, adoption for OpenCode, a totally open-source AI coding agent without any company-sponsored efforts to promote market awareness, already reached 7%, with 42% awareness.
The stickiness number is arguably the most informative of all. Claude Code isn’t just widely used, but is in fact the primary development tool for 31% of developers. That represents what the survey report calls “an almost 80% conversion rate” from regular use to primary tool. Those who try it tend to dive in and go all in.
The Terminal Beat the IDE
Why would a command line be preferable to a sophisticated editor experience? Because the nature of the work has changed. An auto-completion plug-in suggests the next line as you type.
A terminal agent handles the entire ticket: accessing and modifying source code, running tests, and opening a change request with no interruption of work.
The consequence is a major shift in where development happens: The editor is no longer the locus of development activity.
The Standardization Question
For the independent developer paying for a single AI seat and wondering whether they’re using the same tools as their peers, this offers the first large, independent look at where the market has converged rather than where capital has been invested.
The same is true for the small agency trying to establish a standard across three contractors or the sole practitioner wondering whether the era of “vibe coding” has produced a default tool.
It has. And the answer isn’t the incumbent.
This is a completely different market than that described by The Register in May, when its survey coverage found that web developers were using AI on a daily basis with quiet concern that the technology might eventually replace them. Three months later, the anxiety hasn’t gone anywhere, but the debate over whether to use AI tooling appeared to be over. The decision had already been made.
Hosts Are Already Rebuilding for This
The finding most relevant to hosting providers is that 90% of developers now use agents on a weekly basis. As a result, the actor initiating actions withint he deployment process is increasingly an agent operating from a terminal, rather than a person working in an editor.
Developers using agents on a weekly basis
That has important consequences. Deployment paths, formats of build logs, types of error messages, and methods of enforcing quality control must all be interpretable for an agent — predictable, structured, and documented clearly enough to support automated action.
A human developer can wrestle with interpreting a cryptic build failure, infer what went wrong, and search for a solution on the internet. An agent has less room for interpretation. When something goes wrong, the system needs to communicate what happened and, ideally, what action should come next.
And to their credit, hosts have recognized the implications. The recent flurry of product launches across the industry is, in many ways, an effort to retrofit the industry for a customer that isn't human. Agent-ready deployment paths, hosted control-panel access to MCP endpoints, and flat-rate application hosting designed for AI-generated products all point to the same thing: The industry's response is not only underway, but it's appropriate.
There remain, however, unanswered questions. Will Copilot's slide stabilize in response to Microsoft's efforts? Will the unbridled growth of OpenCode continue unabated? And will the terminal agent's apparent victory survive the next change in interface technology?
For now, though, the results are clear. Based on a sample of 15,000 developers, the market has settled on a new default. Whether that default lasts is a different question, and history suggests Microsoft would be wise not to count on it.
