AI Speeds Shipping, But Taste Wins: Linear CTO on Quality

AI Lowers Barriers, Amplifying Old Pitfalls

Tuomas Artman, CTO and cofounder of Linear, warns that AI agents like Claude remove engineering friction, making it too easy to ship every feature request or whim. This echoes Steve Jobs' philosophy: "Great products come out of saying no to 999 things and yes to one thing." Without gates, products become convoluted, confusing users. Artman draws from Uber hypergrowth, where relentless shipping outpaced rivals but eroded quality as revenue metrics overshadowed polish. Today, solo AI builders compete with teams, heightening the need for 'tasteful software'—high-quality experiences that provide a moat.

Gergely Orosz, interviewer and former Uber colleague, challenges if this is new; feature factories predated AI. Artman agrees but sees AI democratizing speed, forcing differentiation via craft. At Linear, they reject prototypes, grouping customer requests to solve root problems rather than surface symptoms. AI aids by summarizing feedback, but human judgment crafts ideal UX.

"The pendulum has swung too far into the wrong direction where if you get a feature request you might now be in the position to just immediately ship it and that might be the wrong thing to do," Artman says.

Quality as Competitive Edge Over Time

Metrics like Uber's revenue, trips taken, and time-to-first-trip fail to capture quality until competitors match features. Early Uber engineers obsessed over pixels—Artman recalls his first PR rejected for a two-pixel map overlay offset, measured precisely by the first iOS engineer. This upheld performance, but scale and revenue pressure shifted priorities. Low-price features like Uber Pool boosted metrics short-term, ignoring UX until Lyft matched and users defected gradually to smoother alternatives.

Artman predicts AI accelerates this: ship fast, match features, then lose to superior feel. Linear invests upfront in taste, using AI selectively. Bugs flow constantly; 10% now auto-fixed via single-shot agents creating PRs. Artman envisions near-100% automation soon, freeing humans for design. He critiques Claude Code—Anthropic's tool, reportedly all Claude-built—as buggy despite speed, a symptom of AI arms-race shipping.

"Over time people will pick the one that is of higher quality... it'll just happen over time. There will be no A/B test," Artman explains.

Quality Wednesdays: Cultivating Obsession

Artman's signature ritual started at an offsite: auditing one menu revealed 35 issues, from missing hover highlights (instant on, 150ms fade-out for smoothness) to regressions. The app felt fast via micro-interactions, but lapses accumulated. Team fixed 2,500-3,000 such details since. Now weekly, all 25 remote engineers share one self-found fix in 30-40 minutes—from one-pixel tweaks to backend efficiencies.

Key: Engineers hunt proactively for Wednesdays, embedding vigilance into daily work. Unrelated features get polished en route, slashing regressions. Orosz calls it aspirational; Artman urges all teams, especially with AI easing hunts.

"If you think about quality all the time... you're bound to make less mistakes," Artman notes.

Zero Bug Policy: Immediate Accountability

Bugs accrue constantly; backlogs balloon until crisis triage matches inflow—two months late. Linear's fix: three weeks halting features to zero the queue, then enforce. Agents auto-assign by code ownership; highest priority. Fix same-day (often 2-3 hours) or triage low-impact ones. Users love rapid resolutions—email: "Refresh, it's fixed."

Bugs ≠ Quality Wednesdays (proactive polish). With AI pinpointing issues, every company should adopt: constant fix rate means zero policy trades nothing for perfection.

"There's a very small trade-off... all you need to do is stop development of new features for as long as it takes," Artman advises.

AI's Blind Spots: No Taste, No Feel

AI excels at code, tests, even animations—but lacks 'taste.' It generates functional UIs without perceiving time (e.g., 2s click feels slow?), spatial harmony, or emotional flow. Linear design engineer Emil's X demo: agents built pop-ups/button highlights competently (ease-in curves), but manual tweaks made them 'natural.' AI is timeless, screenshot/DOM-bound; no frustration from lag.

Artman: Hand rote tasks (bugs) to agents; humans own UX judgment. Future tasteful AI? Possible last bastion.

"They have no taste... they simply don't," Artman states bluntly.

Key Takeaways

• Say no to 90% of requests: Group feedback, solve roots, design thoughtfully—AI summarizes, humans decide.
• Implement Zero Bug Policy: Auto-assign, fix immediately (or triage); halt features briefly to zero backlog—users rave.
• Run Quality Wednesdays: Mandate weekly self-found fixes, share in 30 mins—builds product-wide vigilance.
• Obsess pixels and feel: Instant highlights, 150ms fades; measure what revenue misses.
• Use AI for grind (10%+ bugs auto-fixed), not craft—leverage speed without sacrificing taste.
• Watch competitors: Match features lose to gradual quality wins—no A/B needed.
• Proactive polish during features: Wednesday hunts train constant awareness.
• Critique tools ruthlessly: Claude Code buggy from haste—quality signals maturity.