Linear moves sideways to agentic AI as CEO declares issue tracking dead
Linear, the project management platform backed by $82 million in a June 2025 Series C at a $1.25 billion valuation, launched its Linear Agent into public beta on March 24, 2026 — and immediately declared its own category dead. Issue tracking, the company wrote on linear.app/next, is dead. The tool Linear spent four years building to replace JIRA is now, in the company's own telling, obsolete.
It's a bold reframe from a company that raised at a nine-figure valuation eight months ago. But the launch reveals something more interesting than the pivot itself: Linear has built an MCP server that lets external AI agents like Claude and Codex read and write Linear data. The Linear Agent itself, however, has no stated inbound MCP interface — a gap customers flagged within hours in Linear's own Slack channel, where users asked when the Linear Agent would expose MCP so other systems could query it. An employee responded that the team is working on it.
That apparent gap — an outbound MCP server with no stated inbound interface for the Linear Agent itself — is the real infrastructure story here.
MCP, the Model Context Protocol Anthropic open-sourced in late 2024, has become the de facto standard for connecting AI agents to external tools and data sources. Linear's decision to build an MCP server for inbound data access was sensible: it let the company participate in the agent ecosystem without waiting for every agent framework to build a Linear integration. But leaving the Linear Agent itself without an MCP interface means the agent is a walled garden. It can read what's in Linear. Other agents can read what's in Linear. But nobody — not Codex, not Claude, not any MCP-compatible system — can hand a task to the Linear Agent and get a result back through the same interface.
This is a deliberate architectural choice, and it's one with strategic implications. If the Linear Agent had an MCP interface, it would become composable: a Codex instance could dispatch a Linear Agent task as one step in a larger agent graph. Without it, Linear controls the agent and the data, and customers who want agentic Linear functionality are locked into Linear's own UX. The company that built an open integration layer for inbound data access has built a closed execution layer for its own agent.
The Register flagged the security dimension: Linear Agent is on by default and operates within existing user permissions, meaning it inherits whatever access the logged-in user has. Generative AI systems can be vulnerable to prompt injection, and a default-on agent with broad access permissions is a different risk profile than a feature users opt into. Linear's documentation says workspace admins can disable it in Settings > AI > Linear Agent, but that's a settings change — not a permission model. The agent runs under the user's identity, not its own scoped identity. This is the same over-privilege pattern documented across the enterprise agent landscape in recent CSA and Aembit research: agents inheriting human permissions because that's the path of least resistance, not because it's the right architecture.
Linear's own numbers, which the company is treating as evidence of the pivot's necessity, come with the standard self-reporting caveats. Coding agents are installed in more than 75 percent of Linear enterprise workspaces, the company said on linear.app/next. The volume of work completed by agents grew fivefold in the last three months. Agents authored nearly 25 percent of new issues in Linear workspaces. These are Linear's own figures, based on Linear's own data, offered without independent verification. That doesn't make them false — agent adoption is visibly real in the developer tooling space — but it means readers should weight them accordingly.
The comparison worth making is Basecamp, the project management tool from 37signals, which The Register noted is repositioning itself as agent-first and agent-native. Basecamp's approach and Linear's are different in one important respect: Basecamp is building infrastructure it expects the broader agent ecosystem to use. Linear is building an agent that runs inside Linear's own product. The question neither company has fully answered is what happens when customers prefer agents that compose across tools over agents that are contained within one.
Pricing during the beta period is zero — all Linear Agent features are available at no additional cost, with usage-based pricing expected at general availability. Automations and Code Intelligence require a Business or Enterprise plan. The MCP server is a separate product with its own economics, which Linear has not yet disclosed.
For now, the most honest framing of what Linear announced is not an agent revolution but an agent addition. The Linear Agent is real shipped code, not a concept or a roadmap. It's on by default, it's doing work, and 25 percent of new issues in some workspaces are coming from it. Those are real signals. But the MCP asymmetry tells you where Linear's strategic priority actually sits: the company wants to be the system agents query, not the system that queries other agents. Whether that's a durable position or a temporary one depends on whether the rest of the agent ecosystem agrees that staying inside Linear's walls is the right architecture.
Primary sources: The Register | Linear's changelog announcement | Linear's next page | Linear Agent documentation | Sacra on Linear's Series C | Linear's MCP server