---
title: 'In the News: August 19, 2026, Extra 2'
description: 'Linear published two years of its own workflow data: teams with a coding agent went from 21 to 65 pull requests a week, and the time they spent did not fall.'
canonical_url: 'https://darkfactory.dev/news/2026-08-19-extra-2'
markdown_url: 'https://darkfactory.dev/news/2026-08-19-extra-2.md'
collection: news
date_published: '2026-08-19T10:30:00-04:00'
date_modified: '2026-08-19T10:30:00-04:00'
---

# In the News: August 19, 2026, Extra 2


Linear's two years of aggregated product data show that weekly pull requests roughly
tripled for teams that connected a coding agent, while the time those teams spent
inside Linear did not fall.

## 1. Linear's own numbers show output up and hours not down

**[AI usage patterns in software teams](https://linear.app/data)** · Tim Qi, head of data, Linear · How teams build, Edition 01, 2026

The report draws on aggregated product data from Linear's paid workspaces back to June 2024, and cuts it three ways: who adopts AI features, where time goes, and how much ships. On a fixed cohort of 6,887 paid teams, 4,280 with a coding agent connected and 2,607 without, weekly pull requests opened per workspace went from 21 to 65 for the agent teams between the weeks of June 2, 2024 and June 21, 2026. Teams without an agent went from 8 to 10 across the same span. Over all 47,900 paid workspaces, pull requests per workspace per week are up 111% on the June 2024 baseline, roughly flat for the first year and bending upward from early 2026. Linear says the two cohorts are not directly comparable, because the agent teams were already shipping more before coding agents existed.

The time series is where the report cuts against the usual sales case. Comparing June 2025 with June 2026, across 54,300 and then 89,000 paid users, average minutes per user per month spent creating and triaging issues rose for engineering from 24 to 28, and commenting from 35 to 40. Planning barely moved: customer requests, docs and projects all sit within a minute of where they were. Two categories that did not exist a year earlier now show up in every function's month, chatting with AI and delegating issues to agents, with product logging five minutes a month on AI chat against zero the year before. The report's reading of that: "Nothing else shrank to make room, which suggests AI has landed on top of existing work rather than replacing any of it, at least so far." Its closing note is blunter. "As far as we can observe, teams are working more, not less, suggesting AI has a Jevons paradox quality beyond token consumption."

Who files the work has shifted too. In the week of August 3, 2026, agents and MCP clients created 2,435 issues against 2,481 from people and integrations, excluding imported issues. In the week of June 3, 2024 the same two counts were 0 and 605. The share of product managers attaching a pull request in a trailing 30 days went from 3% to 10% between June 2024 and June 2026, and designers from 1% to 8%, measured on 166,000 paid users. Among 13,300 executives, CEOs at companies of 201 or more employees went from 9% to 36% active on AI features between January and June 2026, the largest jump of any cut in the report.

The caveats are the report's own, and it does not bury them. This is one vendor's customer base, stated plainly: "this is a picture of adoption inside our own customer base, not the market at large." Pull requests are counted opened rather than merged, and only in repositories connected to Linear, which the report calls a floor rather than a ceiling. Company size comes from third-party enrichment, so that cut covers fewer workspaces than the rest. The measure itself gets conceded: "Many will rightfully argue that looking at pull requests indicates motion rather than value, which is certainly true, but it's still a step forward from measuring tokens." On business results the claim is limited to correlation. "We have no way of knowing whether this increased output led to positive business outcomes, but it shows a very clear correlation between AI adoption and acceleration."

**Why it matters:** most of what a practitioner can cite about agent throughput is either a controlled study on short tasks or one person's account of one repository. This is neither. It runs two years and it publishes its own denominators. If you are writing a business case that promises hours back, this dataset will not carry it. Output went up and the clock did not come down, which is a capacity argument rather than an efficiency one, and the two get budgeted very differently. Linear argues in the same breath that using token spend as a proxy for value "will be remembered as a relic of AI's early days", while admitting pull requests opened are a weak proxy too. That leaves the number you can defend in a planning meeting smaller than either side of the argument would like.