Taste & Craft: A Conversation with Tuomas Artman, CTO Linear & Gergely Orosz, @pragmaticengineer
TL;DR
AI makes it dangerously easy to say yes to every feature request — Tuomas Artman argues the old constraint was that engineering was hard, but with agents making shipping cheap, teams risk building bloated, confusing products unless they keep saying no like Steve Jobs’ “999 things.”
Linear’s advantage is taste, not raw shipping speed — Artman says competition now includes tiny teams or even one person using AI, so the moat becomes “tasteful software” and high-quality UX rather than matching features.
At Linear, AI is already fixing real bugs autonomously — around 10% of incoming bugs are automatically turned into PRs and merged without engineer intervention, and Artman expects that number to move much closer to 100% over the next few years.
Quality is hard to measure in dashboards, which is why companies neglect it until users slowly leave — drawing on Uber, Artman says revenue and trips taken were easy to optimize, but product quality only showed up later as a gradual loss to competitors like Lyft.
“Quality Wednesdays” turned craftsmanship into a team habit — after finding 35 issues in one tiny UI menu, Linear made every engineer bring one self-discovered quality fix each week, leading to roughly 2,500–3,000 small improvements and fewer regressions overall.
Linear’s zero-bug policy is brutally simple: every bug gets immediate attention — the company paused new feature work for about three weeks to get the backlog to zero, and now most bugs are fixed within 7 days, often in 2–3 hours, which Artman says delights users who report an issue and get a fix email the same day.
The Breakdown
The real risk of AI isn’t slow teams — it’s tasteless software
The conversation opens with Tuomas Artman pushing on a trend he thinks is heading the wrong way: agents make it trivial to ship every idea immediately, and that’s exactly the problem. He reaches for Steve Jobs’ old line about saying no to 999 things, arguing that AI has made “yes” too cheap, which could leave software more convoluted and less usable.
Uber taught him what hypergrowth does to product quality
Gergely Orosz challenges whether feature bloat really started with AI, and Artman says no — he saw the same pattern at Uber during the winner-take-all race against Lyft. In that environment, everything centered on speed, scale, and revenue, and the parallel to AI is that now every company is effectively in a race against much smaller teams who can ship just as fast.
What Linear speeds up — and what it refuses to rush
Artman says Linear absolutely wants to move faster, but not by shipping raw customer requests as-is. The real work is still talking to users, grouping requests into root problems, and designing the right UX; AI helps summarize and cluster feedback, but it doesn’t decide what the product should be. One place it does help a lot: bug fixing, where 10% of reported bugs already get auto-fixed by a single AI run and merged without human effort.
Claude Code is impressive — and visibly rushed
Asked directly about Claude Code, Artman says the product itself reveals the tradeoff Anthropic made. If “all of the functionality” was built by Claude, he says, you can spot bugs and quality problems within seconds, which to him is evidence of a company moving fast under intense OpenAI pressure and letting quality slip.
Why quality dies in metric-driven companies
The Uber story gets more concrete here: the company had five big metrics, but revenue was the one everyone really chased. Artman remembers an early PR where an original iOS engineer rejected his change because a UI element was off by two pixels — a tiny thing no user would consciously notice, but proof that craftsmanship existed before scale and incentives crushed it.
Quality Wednesdays: the ritual that made everyone care
One of Linear’s signature practices came from Artman’s frustration that engineers were missing subtle UX details, like hover states needing to feel instantaneous on the way in and smooth over 150 milliseconds on the way out. In one offsite exercise, the team inspected a single small menu and found 35 issues; that shock led to weekly meetings where every engineer brings one quality improvement they found themselves, a practice that has now produced around 2,500–3,000 fixes.
Zero-bug policy means bugs stop everything
At Linear, bugs don’t go into a graveyard backlog. A newly reported bug is automatically assigned to the person most likely responsible, becomes their top priority, and usually gets fixed in hours; not every bug must be fixed, but every bug must be decided immediately. Artman says the company spent about three weeks pausing new feature work to reach zero, and now users routinely report something and hear back two hours later that it’s fixed.
AI still has no taste — and engineers need more product sense
Artman’s bluntest line is that AI “doesn’t feel” and “has no taste.” It can generate UI and animations, but it can’t sense frustration, judge timing, or know why one motion feels natural and another feels weird; he points to a post by Linear design engineer Emil showing agent-made animations versus manually refined ones, where both worked but only one felt right. Looking ahead a year, he expects software engineers to become more like product engineers — closer to customers, more responsible for what should be built, and less valued for pure implementation alone.