Copilot is now the minority tool, and your 2025 standardization decision knows it

Awareness is not the variable doing the work here. Claude Code and Copilot are known to the same 79% of respondents ^[8], so both draw from an identical pool of people who have heard of them. Convert it: 39 points of workplace use against 79 points of awareness puts Claude Code at 49% of the developers who know it exists, while Copilot converts 27% ^[3]. Codex turns 65% awareness into 16 points of use, about a quarter; Cursor manages 16% of its 75% ^[4]. Cursor's awareness went up while its workplace use went down ^[10], which is the shape of a tool people tried and then left out of their working set.

The renewal arithmetic is what should worry anyone holding a signed agreement. A Copilot standard set in mid-2025 was set on a number that had been sitting between 29% and 31% for months ^[7], and stability is precisely what a standardization committee buys. Twenty-one percent is nine points off that band's midpoint, a 30% relative decline, and the gap between the last stable reading and the new one is roughly five months ^[2].

Now add the four named tools: 39 plus 21 plus 16 plus 12 gives 88 points of workplace adoption against the 90% who use an agent weekly ^[5]. Since the same survey has open-source agents and other commercial products gaining ground too ^[14], that sum only balances if a good number of developers are running more than one tool at work. The standard is already plural at the desk, whatever the master agreement says.

All of this arrives second-hand. The figures are a dev.to analysis of JetBrains' data, and the same post pauses mid-argument to recommend an API gateway for organisations managing multiple model endpoints ^[12]. The headline number is written as "up to 90%" ^[3], which is a hedge, and the question wording behind it is not on the page. Weekly use of an agent is a low bar: one prompt on a Tuesday clears it, and the survey's own description of what agents do, reading whole repositories and running test suites ^[13], is a much higher one.

What the material supports is narrower and more useful than the rank order. The tool many organisations blessed about a year ago is now reported in use at work by one developer in five ^[7]; the one they did not bless is used by two in five, roughly 1.9 times as many ^[1], and 47% in the United States ^[6]. A procurement decision does not have to have been wrong to be stale. The cost of reopening it is a week of arguing, against a year of paying for seats that people are routing around.