Linear's Data Shows AI Now Writes Half of All Issues — and Teams Are Working More, Not Less

Two years ago, fewer than one Linear issue in a thousand was created by AI. Today, agents and MCP clients author just under half of everything created inside one of the most widely used project trackers in software — and at the current pace, they will soon write more issues than people and integrations combined.

That is the headline finding of “How Teams Build,” the first data report from Linear, published this week as Edition 01 under Head of Data Tim Qi. Unlike the token-usage dashboards and code-volume metrics that model vendors and coding-tool makers like to publish, Linear’s dataset captures an entire product-development workflow — from the first issue through triage, comments, and the pull request that closes it. The result is one of the sharpest independent pictures yet of what AI assistance actually does to software teams in 2026. The short version: output is up, roles are blurring, and nobody is saving time.

Agents now author half the work

The most striking chart in the report tracks issue creation by source from June 2024 to August 2026. The agent-and-MCP line sits at effectively zero through the end of 2024, flickers to life in the spring of 2025, and then goes vertical: from roughly 200,000 AI-authored issues per week in January 2026 to about 2.4 million per week by early August 2026. Human- and integration-created issues grew too — from around 600,000 to about 2.5 million weekly over two years — but the agent curve is on track to overtake it.

Importantly, this is not a niche sample. Linear says tens of thousands of teams build software inside the product every day, and the report draws on paid workspaces only: 127,000 users active in both January and June 2026 for the adoption analysis, 199,000 users for company-size cuts, and 47,900 workspaces for the pull-request trend. The company is candid about the limits — it cannot see AI usage that happens outside Linear, so this is a picture of adoption within its own customer base, not the market at large.

Everyone became a builder — including the CEO

Adoption of Linear’s AI features more than doubled in every job function between January and June 2026. Product managers climbed fastest, from 12% to 34%. Engineering went from 12% to 30%. Even go-to-market — the function furthest from the codebase — jumped from 5% to 18%.

The executive cuts are more surprising. CEOs at companies with more than 200 employees went from 9% to 36% AI-active in six months, the single largest jump in the entire report. CTOs at those companies reached 35%, CPOs 24%. Linear’s read: the most senior leaders are learning the technology by using it, not by reading about it. Company size, usually a strong predictor of how fast organizations adopt new tools, “barely registers” — adoption roughly tripled at startups and enterprises alike.

That role-blurring extends to shipping code. The share of product managers who attached a pull request in the last 30 days rose from 3% to 10% in two years; designers went from 1% to 8%. The people who used to describe a change increasingly ship it themselves — and since Linear only counts PRs in connected repositories, those numbers are floors, not ceilings.

Coding agents tripled output

On the output side, pull requests opened per workspace are up 111% against a June 2024 baseline. Output held roughly level for the first year, then bent sharply upward through 2026 as model quality and adoption climbed together.

The cleanest story comes from a fixed-cohort comparison of 6,887 paid teams: workspaces that connected a coding agent roughly tripled their weekly pull requests over two years, from 21 to 65, while teams without an agent went from 8 to 10. The agent teams were already higher-output before agents existed, so the levels aren’t directly comparable — but each cohort measured against its own baseline leaves little ambiguity about where the acceleration lives.

The catch: time spent went up, not down

Here is the finding that will sting anyone who bought AI tooling on a productivity promise. Time spent creating, triaging, assigning, and commenting on issues rose in nearly every function between June 2025 and June 2026. Founders show the biggest swings: up 17 minutes per month on create-and-triage and up 26 minutes on commenting. AI chat and agent delegation — categories of work that did not exist a year ago — now appear in every function’s week, and crucially, nothing else shrank to make room. Planning time in Linear barely moved at all.

Linear’s own conclusion is blunt: “the overall time spent on product development is going up rather than down. As far as we can observe, teams are working more, not less, suggesting AI has a Jevons paradox quality beyond token consumption.” Efficiency gains in the tool get reinvested as more total work — more issues, more coordination, more pull requests — rather than returned as leisure or headcount savings.

Motion, not value

The report is refreshingly honest about what it cannot say. An opened pull request “says nothing about the value of the change.” Linear has “no way of knowing whether this increased output led to positive business outcomes.” The authors argue that PRs are still a better proxy than token spend — a mechanical refactor can burn enormous token budgets while a meaningful bug fix doesn’t, so using tokens as a value proxy “will be remembered as a relic of AI’s early days.” Future editions will try to trace the full lifecycle from token spend to outcomes, which Linear says it can newly observe now that code review runs through the product too.

For engineering leaders, the practical takeaways are uncomfortable but useful. First, AI adoption is no longer an engineering-local phenomenon — it is org-wide, executive-included, and size-independent. Second, if your ROI model assumed time savings, the data says the opposite so far: expect more throughput, more coordination overhead, and more total hours. Third, the teams compounding fastest are the ones that wired coding agents into their workflow early. The acceleration is real; the productivity dividend is still pending.