Linear COO: Agents Create Over Half of All Issues in a $100M ARR Product
On this page 7 sections
TL;DR
Linear's COO Cristina Cordova gave Latent Space 84 minutes on the go-to-market of a company whose product now files most of its own tickets: agents create more than half of all issues in Linear, up from 3% a year ago. The business passed $100M in ARR this year, has 40,000-plus paying customers at 177% net revenue retention, and employs roughly 170 people. It has never hired an SDR, it prices agent coding at the published API rate plus a small markup, and its own CEO declared its original category dead in March.
Half the tickets write themselves
When CEO Karri Saarinen published "Issue tracking is dead" on March 24, agents were authoring nearly 25% of new issues, agent-completed work had grown 5x in three months, and coding agents ran in more than 75% of Linear's enterprise workspaces. By the company's August 26 tender post, agents sat in 95% of paid workspaces and created 50% of the work. Cordova updated the number again in the interview: "over 50% of issues created on Linear are created by agents."
The change reaches the code. Issues with a pull request attached by someone in an engineering, product or design role grew sevenfold since the start of 2026, and Linear's coding sessions workflow now resolves roughly 30% of the company's own incoming bug reports, mostly on the first pass. Linear shipped its MCP server in May 2025, which Cordova calls "basically the first company in our space to do it."
Her warning for everyone else selling software right now: do not settle for being the database behind that MCP server. "I think you have to bring the agentic workflows into your tools and you have to make them better than the tools that people would be using externally to interoperate with your system. Otherwise, there's no reason for people to go into your tool." A database with an MCP endpoint is a kitchen that only takes delivery orders: the food still moves, and the customer belongs to the app. So Linear watches two lines, agent usage and app usage, and both are climbing, because a chat thread is a poor place to see a whole team's work.
Charge for what it costs you
Seat pricing survives. Seats were the entire bill until June, when coding sessions split it. Linear Agent chat stays free even though it runs on frontier models. In a coding session the user picks an OpenAI or Anthropic model and harness, and Linear passes the cost through: "here's what we were charged from the API... and we charge like a little bit on top of that."
The pricing model she admires and cannot copy is Intercom Fin's $0.99 per resolved conversation. Outcome pricing fails for triage because the outcome has no stable value: one company thinks routing bugs into queues is the PM's job, another wants its PM talking to customers. Until someone cracks that, Linear charges for what the tokens cost it and keeps the rest free.
170 people and no SDRs
Cordova joined in 2023 at about 40 people, with one salesperson on staff, and three weeks in she crossed "move upmarket to enterprise" off the company goals in front of the whole company. She aimed at growth-stage companies instead, the rip-and-replace crowd, and the business is at least 10x since. The enterprise segment only started in 2025, and its tidy roadmap got eaten: the MCP server, the agent platform APIs, and the Codex, Cursor and GitHub Copilot integrations were all unplanned, and much of the enterprise feature list never shipped.
The sales org still has zero SDRs, because a qualification call bills the prospect's time to feed your pipeline: "we don't have any SDRs." Account executives run their own outbound with tooling that keeps it personal. There are no forward deployed engineers either; solutions engineers sell alongside enterprise AEs and sometimes build a working ServiceNow integration overnight to prove a migration is possible.
Support is in-house and never lets a model answer a ticket alone. AI drafts, a human decides what goes out, and customers sometimes ask whether the reply was AI because it came back so fast. The suspicion is fair; a model wrote most of it, and a person signed it. In three and a half years she has had zero escalations over a bad support experience.
Headcount discipline comes from a rule she brought from Stripe: at least 40% of the company in engineering, product, design and data, and if sales wants more heads it helps recruit engineers first. At about 170 people, the entire C-suite is the founders and her.
The AI budget is doing the selling
What unblocks buyers at older companies now is the budget line, and it is fat: "for some companies that is maybe not infinite, but it is very high." Anything that fits inside the AI budget skips the procurement fight a plain productivity tool loses.
Contracts are shortening at the same time. Buyers sign one-year deals and paid pilots because nobody can predict what they will want in two years, and a multi-year lock means no price moves without step-ups written in. Cordova's read on the overnight ARR darlings is the sharpest part of the interview: "you sign a few deals that are, like, 30 to $40,000,000 in ARR. And then, okay. Great. We got to a 100,000,000 ARR." A defense contractor or a giant law firm can hand you that in two signatures, and 40,000 customers on seats is a different business. When those one-year deals hit renewal, everyone finds out whose ARR was real.
What it means if you build
- Instrument two lines separately: work created by agents and work created in the app. If the agent line climbs while the app line flattens, your product is collapsing into your API.
- When intelligence is a pass-through cost, price it like one. The published API rate plus a small markup beats guessing a margin on tokens you rent by the minute.
- Free beta with notice, then charge, starting high. Repricing upward is harder than drifting down, and paywalling a formerly free feature spends trust you will want at renewal.
- AI drafts with a human decision point scales support without outsourcing it. Linear ran that way through a 10x revenue run with zero support escalations.
- Agent infrastructure was worth more than the tidy enterprise roadmap. Competitors waited to ship MCP and agents; enterprises shortlisted Linear anyway.
Key Takeaways
- Agents now create more than half of all issues in Linear, up from 3% a year ago; agents sit in 95% of paid workspaces, and issues with an attached pull request grew sevenfold since the start of 2026.
- The financial base: $100M ARR passed this year, 40,000-plus paying customers, 177% net revenue retention, and an August tender that sold $99M of employee stock at a $2.5B valuation in a cashflow-positive company.
- Pricing after June: seats remain, Linear Agent chat is free on frontier models, coding sessions bill the published OpenAI or Anthropic API rate plus a small markup, and Loops run on AI credits.
- GTM at $100M ARR: about 170 people, zero SDRs ever, no forward deployed engineers, support in-house with AI drafts and human sign-off, and the founders plus the COO as the whole C-suite.
- AI budgets speed purchases up and shorten them: one-year deals and paid pilots are the default, and some fast $100M ARR runs rest on a handful of $30-40M contracts with defense or legal buyers.
- "Issue tracking is dead" came from Linear's own CEO in March: the dead part is humans manually moving work through a system, and the replacement pitch is context turned into execution by agents.
Sources: Latent Space, interview with Cristina Cordova, Nov 19, 2026, Linear, Sharing growth with the people building it, Aug 26, 2026, Linear, Issue tracking is dead, Mar 24, 2026, Linear changelog, Coding sessions, Jun 11, 2026, Linear changelog, Introducing Linear Agent, Linear changelog, Introducing Loops, Linear changelog, MCP server, Intercom, Fin AI Agent outcomes pricing