Research
AI Minute Newsroom
2026-08-19
Almost half the tasks in a large project tracker are now written by software rather than by a person
Linear, the issue tracker used by a large slice of the software industry, opened its own numbers this week — a public dashboard drawn from about 47,900 paying workspaces and covering June 2024 to August 2026. The headline figure is that agents now create roughly as many issues per week as humans and integrations combined: around 2.44 million against 2.48 million, a line that has almost converged. The effect on output is uneven in a way worth noticing. Teams that have adopted coding agents went from about 21 pull requests a week to 65; teams that have not went from 8 to 10. Adoption also stopped being an engineering matter: the share of product managers active on AI features rose from 12 to 34 percent in the first half of 2026, go-to-market staff from 5 to 18 percent, and the proportion of product managers who now attach actual code changes to their work went from 3 to 10 percent, designers from 1 to 8. These are Linear's own figures about its own product, self-published and not audited by anyone outside the company.
Why it mattersMost claims about AI productivity come from surveys, where people estimate how much time they think they saved. This is a log file: it counts things that were actually created. What it shows is not a general uplift but a split — the teams that restructured around agents pulled away sharply, and everyone else barely moved. That is a harder finding than a simple 'AI makes you faster', because it suggests the gain is in the workflow rather than in the model. The other number to sit with is the third of product managers now shipping code: the boundary between who specifies work and who does it is dissolving faster inside these tools than most organisation charts admit.
✓ Verified · 2 sources
Read in the app — free, in 9 languages
Related stories
OpenAI will ship a Codex upgrade daily for 28 days or reset limits.
2026-10-05Mathematicians cracked five open problems using an ordinary chat box.
2026-10-05Untuned models solved agent tasks their polished versions could not.
2026-10-04Google paused its open-source bug bounty after a flood of AI reports.
2026-10-04Anthropic is paying to train 10,000 engineers to deploy Claude.
2026-10-04