AI writes 48% of all software issues now. Teams ship slower, not faster.

Linear just published usage data from 127,000 paid workspaces.

AI agents author nearly half of all issues created on the platform.

Up from fewer than 1 in 1,000 two years ago.

The numbers look incredible on paper. Coding agents tripled weekly pull requests. Non-engineers are shipping code — PMs went from 3% to 10% attaching PRs. Designers went from 1% to 8%.

Then Linear showed what happened to delivery.

Total product development time went up. Not down.

This is the Jevons paradox applied to software. AI made it nearly free to start work — writing issues, generating code, drafting solutions. So teams started more work. Exponentially more work.

The marginal cost of beginning something collapsed. The cost of finishing it did not.

Review, validation, integration, and deployment are still bottlenecked by human judgment. AI generates defined work faster than any organization can decide what deserves to ship.

Linear's own data confirms it: teams with coding agents now open 6.5x more weekly PRs than teams without. But PR volume measures motion, not value.

Your engineering metrics are lying to you. Issues created, PRs opened, tokens consumed — all inflating. Cycle time, change-failure rate, rework — the delivery metrics that matter — aren't moving.

Audit your DORA metrics right now. If your AI adoption dashboard shows green while your deployment frequency is flat, you are measuring activity, not output.

SOURCE: https://linear.app/data
VERIFIED: Linear official data dashboard, AI2Work analysis (Aug 20), The Coe Lab report (Aug 19)
SIGNAL: This is the first hard dataset proving AI makes teams busier, not faster. Every executive measuring AI ROI by output volume needs to see this.