Linear CEO: AI Design Tools Should Connect Visual Editing Directly to Coding Agents
Karri Saarinen, CEO of Linear, proposed that his ideal AI design tool should not only generate interfaces but should center around a canvas that directly renders any view of the application. This would allow teams to replicate, modify, complete, and track changes around real interfaces, and then deliver these changes to coding agents in the form of diffs.
He emphasized that the key is not just to "draw faster," but to integrate design language, product constraints, source metadata, modification records, and user feedback into the system, transforming design drafts from static deliverables into queryable, traceable, and executable contextual layers. This aligns with Linear's recent assessment that the bottleneck in software development is shifting from writing code to organizing feedback, decision-making, and context for agents to understand and execute.
The broader industry context is that Figma has opened up agents to work directly on the canvas, using skills to constrain the generation process and supporting self-iteration after screenshots; at the same time, Figma is integrating design context into coding tools to shorten the link from design to code.
Source: Public Information
ABAB AI Insight
Karri's statement points not to "AI will do design," but to the need for a change in the database structure of design tools. In the past, design files primarily served human browsing and handover, with the core assets being pages and components; now, if downstream executors become agents, the system must retain sources, constraints, reasons for changes, and history of differences, because agents require not just an image, but a computable decision context.
This indicates that the design industry is shifting from "interface production" to "intent orchestration." Whoever defines design language, product rules, and diff formats will control the layer between design and development. Figma's introduction of agents into the canvas essentially safeguards the design entry point; Linear, on the other hand, attempts to organize feedback, decisions, and code into workflows consumable by agents. Both are competing for the same thing: contextual sovereignty in software production.
On a deeper level, this also represents a re-slicing of labor division in the software industry. In traditional links, intentions are communicated between product, design, and engineering through documents, tickets, and meetings, leading to significant "translation costs"; once canvas changes can be directly exported as coding plans, many intermediate coordination layers will be compressed, concentrating designers' value on defining constraints, aesthetic judgments, and interaction trade-offs, rather than static draft outputs.
The emergence of such tools now is not only due to stronger models but also because organizations are beginning to recognize that "code implementation" is no longer the most scarce link. Linear's public assertion that "issue tracking is dead" essentially reflects the same trend: execution is becoming increasingly automated, and what is truly scarce is high-quality context, clear intent, and inheritable decision trajectories across teams.