Output Really Did Rise, and the Gap Tracks Agent Adoption
Start with what went up. Two years ago fewer than one in a thousand issues in Linear was created by AI; today it is close to half, and at the current pace it will soon exceed people and integrations combined. On the PR side, pull requests per workspace are up 111% since June 2024 — but that average hides a split. Teams using coding agents went from 21 to 65 PRs per week, a tripling. Teams without them went from 8 to 10, essentially flat. Adoption also stopped being an engineering phenomenon. Between January and June 2026 the share of users active on AI features roughly doubled in every function: product managers 12% to 34% (+22 points), engineering 12% to 30%, design 6% to 22%, go-to-market 5% to 18%. CEOs at organizations of 201+ employees posted the steepest gain at 27 points. Non-engineers are shipping code too: product managers attaching PRs rose from 3% to 10%, designers from 1% to 8%.
The Half That Did Not Fall
In the same dataset, from June 2025 to June 2026, time spent creating, triaging and commenting rose across nearly every function. Engineering added about 5 minutes a month on create and triage alone; founders added 17 minutes on creation and 26 on commenting. A new category appeared as well — "chat with AI" — consuming 2 to 5 minutes a month across functions. Planning work, meaning customer requests and documentation, barely moved. Put the two halves together and the report is not saying AI made development slower. It says that when output rises, coordination cost rises with it: more gets written, so more needs triaging, commenting and aligning. The counterintuitive part is that many teams buy AI tooling expecting coordination time to fall.
Do Not Mix Linear's Numbers with LinearB's
One easy confusion: Linear (linear.app) and LinearB are different companies, both of which published reports recently, and coverage often splices their figures together. Everything above comes from Linear's own product data. LinearB's 2026 benchmark report is a separate sample — roughly 4,800 teams and 8.1 million PRs — finding that 84.5% of manual PRs merge within 30 days against 32.7% of AI-assisted ones, with reviewer pickup time stretching from about 200 minutes to about 1,050. The two point the same direction but measure different things; do not substitute one for the other. For anyone assessing engineering effectiveness, the most usable part of this report is its comparison group. Splitting teams by whether they actually use coding agents produces a 3× gap, not an 11% one. If your PR count has not moved, the problem is probably not which model you picked but whether agents are genuinely wired into the workflow.