AI now authors half of Linear's issues — yet teams ship more, not faster

Linear mined six years of its own product-development data to chart how AI has reshaped software work, and the adoption curve is steep and broad. Between January and June 2026, the share of active AI users at least doubled in every job function — product roles jumped from 12% to 34%, and even go-to-market teams furthest from the codebase climbed from 5% to 18%. The spread ignores company size (adoption roughly tripled from startups to enterprises) and reaches the top of the org chart, with CEOs at larger firms leaping from 9% to 36%. AI now writes just under half of all issues created in Linear, up from fewer than one in a thousand two years ago, and pull requests per workspace are up 111% since mid-2024. Nearly all of that output growth sits with teams that connected a coding agent, which roughly tripled their weekly PRs from 21 to 65 while agent-less teams barely moved from 8 to 10.

The more interesting story is how roles are blurring. Product managers attaching pull requests rose from 3% to 10% and designers from 1% to 8%, so the people who once described a change increasingly ship it, while senior leaders do more hands-on IC work. But none of this showed up as time saved. Coordination work — creating, triaging, commenting — rose across functions, and entirely new activities like chatting with AI and delegating to agents landed on top of existing tasks rather than replacing them. Planning time held flat, suggesting AI has so far changed how teams execute far more than how they decide what to build. Linear frames the net effect as a Jevons-paradox dynamic: cheaper output pulls teams into doing more, not less.

The caveats matter. This is a view inside Linear’s paid customer base, not the market, and any AI work happening elsewhere is invisible. Counts of opened PRs and attached pull requests measure motion, not value — a mechanical refactor and a critical bug fix look the same here, and role classifications lean on normalized job titles and third-party enrichment. Linear concedes it cannot tell whether the output surge produced better business outcomes, only that adoption and acceleration correlate tightly, and it argues that PRs are at least a step up from token counts as a proxy for real work.