Linear, a product development platform "disguised as an issue tracker," is boldly redefining its role in the software development ecosystem by embracing AI agents as first-class citizens. Speaking at the AI Engineer World's Fair in San Francisco, Tom Moor, Head of Engineering at Linear, detailed the company's journey and vision for a future where AI teammates seamlessly integrate into engineering workflows. Linear, already a go-to tool for high-performance teams like OpenAI, Ramp, and Vercel, aims to leverage this new paradigm to tackle long-standing inefficiencies in product development.
Linear's initial foray into AI, starting in early 2023 with a "skunkworks" team, was pragmatic. Despite having no prior AI experience, the team focused on immediate, tangible benefits like summarizing issues and natural language filters. Moor candidly admitted that early attempts at a "Copilot" feature were shelved because "the quality wasn't there... we have this quality bar and it did not reach it." This early restraint, prioritizing subtle utility over flashy but unreliable generative AI, earned them praise from industry observers who noted Linear’s "subtle + useful AI implementations."
However, the landscape shifted dramatically in late 2023 and early 2024. The advent of more capable models like o3, coupled with advancements in multimodal capabilities and vastly expanded context windows, marked a turning point. "The team was reinvigorated by the leap forward," Moor stated, as experiments became "a lot less brittle" and features began to "feel smart." This newfound maturity in AI capabilities allowed Linear to bake deeper intelligence into its platform. They transitioned to a hybrid search approach, combining Turbopuffer with Cohere embeddings, to create "Product Intelligence" – a semantic graph of issues that offers query rewriting, hybrid search, and reranking to provide a comprehensive "map of relationships" between issues and their context.
The ultimate vision for Linear extends beyond intelligent features baked into the core product. Moor envisions "agents as infinitely scalable cloud-based teammates." Recognizing that every team has unique needs, Linear is building a pluggable platform where external AI agents can operate seamlessly within the Linear workspace. Examples like CodeGen, a coding agent that can research best practices and submit pull requests, and Bucket, a feature-flagging agent, demonstrate this potential. These agents, treated as first-class users with identity and audit trails, can interact with issues, create plans, and even perform root cause analysis of bugs. This heralds a future where the prevalent "giant backlog that you're never going to get to the bottom of" will have "no excuse for that anymore," as AI agents tackle grunt work and accelerate development cycles.
Linear's approach to agent interaction emphasizes clarity and transparency. Agents should respond quickly and precisely, reassuring users of their understanding. They are expected to "inhabit the platform" using correct nomenclature and leaving transparent trails of their actions, just like human teammates. Crucially, agents should not try to be "clever"; they must clarify ambiguous requests and default to "helpful silence over unprompted noise," ensuring their contributions are always concise and valuable.
