Linear CEO Karri Saarinen declares 'issue tracking is dead' and reveals Linear's AI-era pivot

Summary

Karri Saarinen, CEO of Linear, published an essay this week arguing that issue tracking is dead — a deliberately provocative framing designed to force software companies to question workflows built for a pre-AI world.

The core argument is that traditional issue tracking was always a bureaucratic workaround. Engineers wouldn't act without a ticket, tickets required negotiation, and the whole process resembled, in Saarinen's words, a line-cook system where someone yells an order and the kitchen eventually starts cooking. Legacy platforms made this worse by encouraging complexity — Saarinen describes prospects arriving with 40-state bug tracking processes and 30 custom fields, asking if Linear can replicate them. His answer is that they shouldn't want to.

AI makes the dysfunction harder to ignore. If an agent can fix a bug in five minutes but the approval and tracking process takes a week, the overhead has become the product. Saarinen wants to compress those workflows down to a single human touchpoint: the agent handles debugging, connects to external tools, proposes a fix, and tags a human only to verify.

Linear's repositioning

Saarinen frames Linear's next category not as issue tracking but as context infrastructure for agents. Linear already aggregates bug reports, customer feedback, project goals, and product state. The argument is that the same context layer that makes human engineers more effective can be surfaced to coding agents so they know what to work on, why, and in what order.

The second piece is agency in the system itself. Historically, tools like Linear were passive — nothing moved unless a human moved it. Saarinen wants Linear to become self-driving: the system detects incoming bugs, starts debugging autonomously, and escalates to a human only at the end.

Linear has already launched a Linear Agent surface for querying that context layer. Saarinen uses it to cross-reference individual feature requests against the broader base of customer feedback. A proprietary coding agent is also in development, framed as a wrapper around existing agents — Codex, Devin, Cursor — but with tighter hooks into Linear's platform that a generic coding agent can't replicate.

Saarinen's sharper point about what AI actually changes in software development: when building anything fast becomes cheap, the bottleneck shifts from execution to deciding what to build. Linear's bet is that the decision layer — understanding, shaping, and prioritising problems — is where the durable value sits.

Why 23% of Linear workspaces still haven't adopted cloud agents

77% of Linear workspaces have connected cloud agents. Saarinen attributes the gap to three things. Cloud agents are still maturing, particularly around environment complexity in large enterprise codebases, though sandbox-based solutions are closing that gap. Timing is a factor for companies still evaluating. The third reason, which Saarinen calls the "stupid" one, is corporate policy: procurement, IT, and security teams are blocking agent adoption through multi-month evaluation cycles, often without a substantive reason beyond institutional inertia.

His suggested fix is a "friction form" — a mechanism where engineers who want access to a new tool file a formal request that escalates to senior leadership with a mandated resolution window of one to two weeks. The logic is that procurement and security teams have a structural incentive not to change things, so forcing a visible SLA with executive visibility changes the calculus.

On enterprise software consolidation

Saarinen is sceptical of two extreme positions circulating in the market. The idea that every tool is converging into the same agent orchestration layer misreads the moment — what looks like convergence is mostly companies noticing the same primitives and trying to figure out who owns which piece of the puzzle. The opposite view, that all existing SaaS should be discarded and rebuilt from first principles, tends to produce rediscovery: teams running ten agents in parallel end up putting them on a Kanban board, then realise they've reinvented a 30-year-old workflow management concept. Some abstractions don't need to be rebuilt.