50% of the Work Created in Linear Is Now Created by Agents. A Year Ago It Was 3%. | Saa Str AI

Overview

AI VC AI Mentor: Digital Jason + Amelia AI Startup Benchmarking

AI Agent Playbook Free e Books

	e Book: Hiring a Great VP of Sales
	e Book: Raising Capital
	e Book:  The First $1m ARR

Details

University All Posts Podcasts The Top CROs VC Fundraising Top Videos Q&A Best of Saa Str #1 Bestselling Book Search Everything Join the Community

Free e Books

	e Book: Hiring a Great VP of Sales
	e Book: Raising Capital
	e Book:  The First $1m ARR

AI Annual 2026 Events Overview Sponsors

	Event Sponsorship
	Media Sponsorship

Digital AI Day 2026 (Free) Speaker Submissions Speaker Requirements Overview

50% of the Work Created in Linear Is Now Created by Agents. A Year Ago It Was 3%.

by Jason Lemkin | Artificial Intelligence (AI), Blog Posts, Saa Str. Ai

Linear published its tender offer news this week:

Buried in the middle of the post is the number every B2B founder should be looking at:

Agents are installed in 95% of paid Linear workspaces, and the share of work agents create went from 3% a year ago to 50% today.

Half the work items in one of the most widely used systems of record in software development are now being created by machines. That happened in twelve months.

I’ve watched a lot of adoption curves over 14 years at Saa Str and as an investor. Almost nothing moves like this inside an established workflow tool.

Two caveats before anyone puts this on a slide. Linear says “the share of work they create,” not “50% of issues,” and we don’t know if that’s weighted by workspace or counted in aggregate across all of them. Installed in 95% of workspaces is an install number, not an engagement number. The 50% figure and the 7x figure are usage, and they point the same direction.

The cohort matters too. Open AI, Cursor, Cognition, Harvey, Physical Intelligence, Legora and Baseten all run product development in Linear. The companies building agents are the ones deploying agents hardest internally, so part of that 50% is coming from the leading edge. Your customers will get there later. They will still get there.

Issues With a Pull Request Attached Grew 7x Since January

Volume is easy to fake. An agent that files 400 tickets a day produces a spam problem and calls it productivity.

Linear’s second data point is that since the start of 2026, issues with a pull request attached by someone in engineering, product, or design grew sevenfold. Work isn’t just being created, it’s being closed with code attached. The loop completes.

If you’re building anything with agents in it, track both: volume created by agents, and completion rate on that volume. The first number can go up every month while your product gets worse.

Atlassian Publishes a Growth Rate Where Linear Publishes a Fraction

So is this a Linear phenomenon or a category phenomenon? I went through what the direct competitors reported in their most recent quarters.

Sources: Linear company blog (Aug 26, 2026); Atlassian Q4 FY26 shareholder letter and earnings release (Aug 6, 2026); monday.com Q2 2026 results and earnings call (Aug 10, 2026); Asana Q1 FY27 results and earnings call (May 2026); Click Up company announcements.

Atlassian’s Q4 FY26 letter has the closest thing to a comparable disclosure. Jira work items and Confluence pages generated through their MCP server are up nearly 4x from the prior quarter. Monthly active users of the MCP server and Teamwork Graph CLI more than doubled in the quarter to surpass 1 million, with total MCP calls up more than 400%. Mike Cannon-Brookes wrote that agents are contributors to the graph, not just consumers of it.

Directionally that’s the same shift. The difference is the shape of the number. Atlassian gave a growth rate off an undisclosed base. Linear gave composition.

Growth rates are the easier number to publish. Up 4x quarter over quarter could be 2% of Jira work items going to 8% and it would be the same headline. Composition can only be published once you’re actually there. Atlassian has every incentive to say “X% of Jira work items are now agent-created” if that number is impressive, and a much larger surface to measure it on. They said 4x instead.

monday and Asana didn’t publish either version. Both reported AI products at 17% of net new ARR, separately and in roughly the same quarter, which is about where AI monetization sits in mid-market work management right now. monday published activity totals instead of composition: more than 1.7 million agent interactions, 98 million AI block actions, roughly 3 million Sidekick conversations and more than 180,000 Notetaker hours since launch. Cumulative-since-launch counts with no denominator.

When a company has the composition number and it’s good, they publish it. Linear did.

98% of Atlassian’s MCP Users Were Also Active in the Jira UI That Month

The fear a year ago was that agents would consume the system of record through an API, the UI would empty out, and seat revenue would go with it.

Atlassian reported that 98% of MCP users were also active in the Jira UI in the same month. Humans and agents working the same records in the same system. Linear’s version of the same finding is the sevenfold growth in issues with a PR attached by a human in engineering, product or design.

Atlassian also put a number on the commercial effect: MCP adopters expand paid seats faster and grow ARR at 2x the rate of non-adopters. Rovo adopters grow ARR more than 2x faster and complete 20% more Jira work items than non-adopters. Two independent companies, one at

Linear held 177% net revenue retention at

177% at that scale is top decile. It means the average account nearly doubled its spend while half the input to the product stopped being human.

At Saa Str we run production ops with 3 humans and 20+ agents. Our own tooling spend went up, not down, when we made that switch. The agents consume more of everything: more API calls, more storage, more records, more compute, more seats in some tools where we had to give agents their own identity. Consumption follows work, and agents do a lot of work.

Atlassian Grew 28% at $6.6B While monday Cut 20% of Its Staff

The AI-disruption story for this category was supposed to be that the incumbent gets eaten. The last quarter showed the opposite.

Atlassian closed FY26 with revenue up 28% to

monday.com grew 22% but guided Q3 to 16-17%, cut roughly 20% of its workforce for about $100 million in annualized savings, and has seen its stock fall 63% over the past year. Asana is guiding to 8-9% growth for the year with a 2-point drag from its PLG motion and bought Stack AI to accelerate its roadmap.

The squeeze is landing on the horizontal mid-market platforms sitting between the incumbent and the fast agent-native tools.

The product gap closed faster than most people realize. Atlassian shipped in one quarter: agents assignable directly inside Jira with access to goals and comment history, Claude and Cursor in Jira (assign an issue and the agent reads it, accesses the repo, opens a draft PR inside Jira’s permissions and audit trail), a Jira Coding Agent that turns work items into ready-to-review PRs, Agent Sessions to track what every agent did and what needs review, and Create with Rovo to spin up context-rich work items.

Linear ships Coding Sessions and Loops. Git Hub ships Agent HQ and a Copilot coding agent that takes an issue and returns a PR. Ramp and Coinbase built their own coding agents into Linear. Click Up acquired Codegen in December 2025.

Issue to pull request is now standard. The differentiator moves to the quality of the context the agent gets and the review surface where a human decides what to do with the output. Atlassian is betting almost entirely on the first, with 200 billion objects in the Teamwork Graph and a claim of up to 44% more accurate answers on 48% fewer tokens when agents are grounded in it. Linear is betting on the second.

One outside calibration: a July 2026 Microsoft study of its own rollout of Claude Code and Copilot CLI across tens of thousands of engineers found adopters merged roughly 24% more pull requests than they otherwise would have, and the authors noted a merged PR is a proxy for output, not for value. Real lift, not a 10x.

They Ran a $99M Tender Instead of Raising a Round

Karri’s framing on the financing is worth copying if you’re in the same position. They didn’t need primary capital, so raising it would have diluted the team and existing shareholders for no reason. Instead they ran secondary so employees could sell part of their vested equity, and got a current mark out of it.

This is becoming the default for capital-efficient companies with real revenue: skip the round, run the tender, let the team take money off the table, reset the 409A and the headline number without adding dilution. Ramp, Figma, and others have done versions of it. If you’re cashflow positive with more cash than you’ve raised, there is no reason to sell more of the company to prove the price went up.

Linear also does equity refreshers and a fully paid month off after four years, then another month every two years after that. And they’re hiring for 30+ roles after years of resisting the standard company-building playbook.

One: what percent of the records in your product were created by an agent last month? Not “do we have an AI feature.” What share of the actual objects, tickets, contacts, tasks, documents, transactions, came from a machine rather than a person clicking a button. Linear could tell the market it went 3% to 50% because they were counting. Atlassian could only give a growth rate.

Two: does an agent have an identity in your product? Linear built an agent platform with agents as first-class actors, rather than a bot account someone shared credentials for. That’s what let them count, permission, and price it. Every B2B product is going to need it.

Three: does your pricing scale with work or with headcount? Linear held 177% NRR through this transition. Atlassian’s MCP adopters expand seats faster than non-adopters. If your revenue is strictly per-human-seat and your customers’ work volume triples while their headcount is flat, you capture none of it.

A workflow metric moved 17x in a year inside a product used by 40,000 paying companies, and the $6.6 billion incumbent is reporting the same shift in its own data. Agents got installed nearly everywhere, they started doing half the work of defining what gets built, and the humans moved to reviewing and shipping it. Two of the best product teams in B2B have already priced that into how they build. Most roadmaps I see have not.

Dear Saa Str: Should I Run 20+ AI Agents the Way Saa Str Does?

New!! Check Out Saa Str's AI Agent Guide. The 20+ AI Agents We Use.

Dear Saa Str: Should I Run 20+ AI Agents the Way Saa Str Does?

New!! Check Out Saa Str's AI Agent Guide. The 20+ AI Agents We Use.

Get from

Key Takeaways

• AI VC AI Mentor: Digital Jason + Amelia AI Startup Benchmarking

• AI Agent Playbook Free e Books

    e Book: Hiring a Great VP of Sales
    e Book: Raising Capital
    e Book:  The First $1m ARR
  

• University All Posts Podcasts The Top CROs VC Fundraising Top Videos Q&A Best of Saa Str #1 Bestselling Book Search Everything Join the Community

• Free e Books

    e Book: Hiring a Great VP of Sales
    e Book: Raising Capital
    e Book:  The First $1m ARR
  

• AI Annual 2026 Events Overview Sponsors

    Event Sponsorship
    Media Sponsorship