How Linear Turned AI Agents Into First-class Users
TL;DR
Linear treated AI as a workflow problem, not a chatbot feature — CEO Karri Saarinen said the team spent “a couple years” resisting the rush to ship generic chatbots, focusing instead on where AI is actually useful inside product work like synthesizing feature requests, triaging bugs, and guiding coding agents.
The company’s big bet is that every team will have many agents, not one — Linear built an open agent platform with strong docs early, which helped it become the default integration point for coding agents from OpenAI Codex to customer-built systems at companies like Coinbase and Ramp.
Linear wants to be the context layer for product development — Saarinen describes the product less as a ticketing system and more as the “backbone” where companies collect signals, decisions, bugs, and intent so agents know what to work on and why.
The metric that matters isn’t token spend or percent of AI-written code — Saarinen mocked the industry’s vanity metrics around PR counts and agent-generated code, arguing that quality indicators like bug volume, user love, revenue, and profit still matter more than raw output.
AI changed Linear’s workflow by collapsing loops, not replacing judgment — The team uses agents to turn Slack conversations into issues, generate first-pass bug fixes, synthesize customer requests, and create shared coding sessions, while still insisting humans take time to find the right problem before moving fast on execution.
Linear is now building its own coding agent because tighter integration beats pure interoperability — Saarinen said third-party agents are useful, but owning part of the agent workflow lets Linear inject context more intelligently, show diffs and previews inside the product, and automate tasks like bug fixing in a shared, multiplayer environment.
The Breakdown
From stealthy craft product to AI-native tool
Dan Shipper opens by framing Linear as one of the rare pre-AI SaaS companies that has actually crossed the chasm well. He recalls using it in 2020 as a kind of secret-software CMS at Every, and connects that same patience and product taste to how the company handled the AI wave.
Why Linear ignored the chatbot gold rush
Karri Saarinen says Linear’s mission never changed: help teams move work forward and remove the overhead of running product development. That’s why the team didn’t rush to bolt on a chatbot when GPT-3 hit; they tried it internally, found it wasn’t genuinely useful, and instead focused on understanding real workflows before shipping anything customer-facing.
The real opportunity: many agents, one system of context
Saarinen’s core view is that companies won’t use one all-powerful agent — they’ll have many, including homegrown ones. That led Linear to build an open agent platform with strong docs, which is why so many coding agents already integrate with it, including OpenAI’s Codex and internal agents at customers like Coinbase and Ramp.
SaaS isn’t dead, but old moats are wobbling
Asked about the “SaaS is dead” narrative, Saarinen gives a more nuanced answer than the market panic. He agrees the uncertainty is real, especially for large public software companies whose inertia used to be a moat, but says the real challenge is to operate like it’s “day one” again and rethink the product around agent-driven workflows instead of clinging to old assumptions.
Inside Linear’s own AI adoption: quality over vanity metrics
Linear has about 120 employees, with roughly 60 on the product team, and Saarinen says nearly all engineers now use coding tools or agents. But he’s openly skeptical of chest-thumping metrics like percentage of AI-written code or merged PR counts; to him, a more honest signal is whether the product is getting better — especially whether bugs are disappearing.
The zero-bug policy and the “find the right problem slowly” philosophy
One of the sharpest moments is Saarinen saying AI agents make him wonder, “why do you even have bugs in your product?” Linear routes every bug into a triage team with a one-week SLA, lets agents do the first fixing pass, and uses that speed only after the company has been deliberate about identifying the right problem — “I want the loop to be fast… but I don’t want the problem finding to be fast.”
Where AI helps today: product synthesis, prototyping, and Slack-to-issue workflows
Saarinen gives practical examples: he built a “Linear way” skill using internal docs and blog posts to synthesize feature requests like multiple assignees or multiple workspaces, surfacing the real underlying need instead of just counting votes. Designers still work manually in Figma because he values the slowness of drawing, but the broader team uses AI to prototype faster, and Slack conversations can now become actionable issues instantly instead of dying in meetings.
The demo: Linear as multiplayer control room for coding agents
The live demo shows where this is headed: custom skills powered by Claude, issue creation from prompts, agent sessions visible to the whole team, code diffs and previews inside Linear, and multiple people collaborating in the same agent thread. Saarinen’s point is that Linear doesn’t need to beat every AI coding tool at everything — it just needs to own the high-leverage upstream workflow where bugs, requests, and product decisions become structured work with the right context.