Linear’s workflow telemetry: more PRs and coordination, not measured requester value

Source. Tim Qi, Head of Data at Linear, “How teams build”, a first-party aggregate study comparing June 2025 with June 2026 activity in paid Linear workspaces, with AI adoption through June 2026 and an issue-creation series through August 3, 2026. The page surfaced in an October 6 search; search date is not observation date. Counts only activity in Linear and pull requests opened against connected repositories; agent teams are workspaces with a connected coding agent, not all teams using one. It gives no individual task ledger or customer outcome.

Three different denominators. In a cohort of 127,000 paid users active both January and June 2026, the 28-day share who interacted at least once with Linear AI climbed for engineering 12%→30%, product 12%→34%, design 6%→22%, GTM 5%→18%; a title-normalization classifier assigns roles. A separate June 2025/June 2026 pair of user populations (54,300→89,000 paid users) yields monthly average minutes in Linear by activity. Engineers' create/triage time rose 24→28 displayed minutes (page reports +5 owing to rounding), commenting 35→40, while customer-request and docs/project time is nearly flat. Agent-issue interaction (+1 minute) and AI chat (+2 minutes) appear as additional visible categories. These are app minutes averaged over all users, not active agent review minutes on a feature; product reports 3→4 rounded customer-request minutes but the printed difference says 0, a warning against treating rounded changes as exact.

Output and role inference. Among 166,000 paid users in June 2026 with connected GitHub repositories, product managers attaching a PR in the last 30 days rose 3% (June 2024)→10% (June 2026), designers 1%→8%; a PR attached is not necessarily written by that role or merged. In 6,887 fixed paid workspaces, those eventually connected to coding agents averaged 21→65 PRs opened per week from June 2024 to June 2026, against 8→10 in other workspaces. The eventual connected group already had a much higher baseline; the observed PR divergence cannot isolate agent causality. Qi explicitly concedes no measured positive business outcome and promises further lifecycle research. Do not translate page dwell into total working time, nor PR creation into completion or customer value.

Contribution to question. The primary vendor data agrees with the planning problem: more active triage/commenting can accompany more opened PRs while time in formal docs/customer-request views barely moves, so decision memory may be hiding in other channels. It partly complicates Dru’s manager-as-builder view with observable executives using AI and product/design users attaching PRs, but neither shows that role shift improves acceptance. It does not close the same-feature time and requester outcome gap. Backlog: weekday argument with Tim Qi’s own closing caveat when the four unread posts clear; ideally pair with a real single-feature human review ledger, not an isolated PR-growth headline.