Bear LiuIdeas, observations and conversations
Product teardowns, podcast conversations and videos about design craft, AI workflows and what actually ships.
All ideas, newest first
14 items
I built an app with Claude Code in 20 days. Most of the work wasn’t code (opens in a new tab)
Twenty days building CUBE, page by page from the paper notebook: what is left for you when AI does the execution.
I asked 3 developers when I was the bottleneck. Here’s what I changed (opens in a new tab)
Three developers, three answers, and the handoff I use now: complete pages first, then complete components.
Hire for judgement, not screenshots: how founders should pick a designer (opens in a new tab)
What actually won a client project, and the small real-problem test I would give any designer before hiring.
What makes 19 million people trust Wise?
Consistency, transparency and what a company built on trust cannot explain when an account is frozen.
MVP vs MLP: why most products stay minimum and never lovable (opens in a new tab)
Taking CUBE from viable to lovable by cutting features first, then fixing how it feels.
Does the double diamond still work when AI can build anything? (opens in a new tab)
Two chapters of a real project testing whether discover, define, develop, deliver still holds.
Why does Google Health keep asking you questions?
How questions collect what passive tracking cannot, and where Google’s confidence ends.
The restraint of Apple Health: how a difficult app helps Apple sell the next Apple Watch
Why a cluttered, passive app is really health infrastructure that keeps you on the iPhone and Apple Watch.
The two keystrokes that made Raycast, and now trap its AI
How two keystrokes became one grammar for 2,500 extensions, and now limit what its AI can do.
Arc never broke through. Why did Dia sell for $610M three months after launch?
What Dia carried over from Arc, and what it left behind, on the way to a $610M sale.
Why Arc was loved by a few but never became a mainstream browser
The wrong audience, too much to learn up front, and a free model. Three problems that beauty could not fix.
How did Linear reach $100M ARR?
Deciding early who it was for, accepting who it would lose, and turning that restraint into $100M ARR.
Free is expensive (opens in a new tab)
Felix Lee, ADPList: vibe coding, why 30% of Figma might disappear, and founders as translators between design and business.
Why selling design keeps designers stuck (opens in a new tab)
Luca Richards, Scarlo: from selling $1,000 websites to taking equity in YC and Antler startups, and why the question became why you.