Taste & Craft: A Conversation with Tuomas Artman, CTO Linear & Gergely Orosz, @pragmaticengineer

In this fireside chat at AI Engineer Europe 2026, Tuomas Artman, Cofounder and CTO of Linear, speaks with Gergely Orosz about craft, quality culture, and the risks of shipping too fast in the age of AI agents. The conversation is wide-ranging but centers on a core thesis: the ability to ship code instantly with AI does not mean you should, and the discipline of saying no to features is more important than ever.

Artman opens by invoking Steve Jobs' maxim that great products come from saying no to 999 things. With AI agents capable of shipping every feature request that arrives, he warns that software quality and user experience will degrade rapidly for teams without strong taste and editorial discipline. He draws a direct analogy to his experience scaling Uber in its hyper-growth years, where revenue metrics drove all decisions and quality was sacrificed until a competitor — Lyft — offered a comparable product with a better experience. His thesis: quality degradation is gradual and never visible in an A/B test, but eventually causes users to slowly defect.

On Claude Code specifically, Artman is candid: he can spot bugs and quality issues in the tool within seconds of using it, and attributes this to Anthropic shipping too fast under competitive pressure from OpenAI. He notes that Anthropic itself claims Claude Code was largely written by Claude — and says it shows. This is offered not as a dismissal but as a cautionary example of the broader pattern.

At Linear, AI has accelerated one dimension clearly: bug fixing. Artman reports that 10% of inbound bugs — whether filed by engineers or customers — are now automatically resolved by a single-shot AI instance, with a PR landed and merged without any human intervention. He expects this to approach 100% over several years.

The talk covers two distinctive Linear practices in depth. First, Quality Wednesdays: every engineer must independently find and fix one quality issue before a weekly 30-minute all-hands review. This practice emerged after Artman held a session at an offsite where the team found 35 problems in a single small view. To date, Linear has fixed approximately 2,500–3,000 such micro-quality issues. The unintended benefit: engineers became permanently alert to quality regressions even while building unrelated features. Second, the zero bug policy: every reported bug is immediately auto-assigned (using agents to identify the likely owner), becomes the assignee's top priority, and must be resolved within seven days — typically in two to three hours. Artman argues the rate of bugs generated is constant regardless of when you fix them, so there is no long-run cost difference between fixing immediately versus accumulating a backlog; the only trade-off is a one-time sprint to drain the existing backlog.

On the question of AI agents and taste, Artman is direct: agents have no sense of time, no ability to feel frustration when a click takes two seconds, and no intuition for what makes an animation feel natural versus mechanical. He cites a recent example from Emil, one of Linear's design engineers, who posted on X demonstrating how an agent-built animation did all the technically correct things but felt wrong — ease-in timing slightly off, motion slightly too fast or slow — while the hand-tuned version felt natural immediately.

Looking ahead one year, Artman predicts that every software engineer will need to become a product engineer: engaging directly with customers, synthesizing feedback, making prioritization decisions, and shipping features end-to-end. He describes Linear's hiring process — including a fully paid one-week work trial — as deliberately filtering for this mindset. For engineers looking to develop product sense without changing roles, his advice is to build something for yourself, ship it, and get real users; and to read Apple's Human Interface Guidelines as the canonical reference for good UX.