Teardowns
Products I didn't build, taken apart.
Each one follows the same frame — the bet, the job, the metric that matters, three decisions, where it's fragile, and what I'd ship next. Facts are sourced and dated; the arguments are mine.
- Cursor / AnysphereJul 28, 2026
Cursor: What $0 → $4B ARR Actually Bought
The fastest-growing SaaS ever measured did not win on model quality — it won on where it chose to sit. A teardown of the wedge, the enterprise handoff, and the one dependency that should keep its PMs up at night.
- DuolingoJul 21, 2026
Duolingo: The Streak Is the Product
56.5M daily actives and 10M people on a 365-day streak. The gamification is the famous part — the interesting part is that Duolingo optimizes a habit metric while selling a learning outcome, and has done it for a decade without the gap catching up.
- Notion AIJul 14, 2026
Notion AI: The Add-On That Had to Disappear
Notion sold AI as a $10 add-on, then killed the add-on and folded it into a $20 plan. That reversal is the most useful packaging lesson available to anyone shipping AI into an existing product — and it wasn't really about pricing.
- SpotifyJul 7, 2026
Discover Weekly: A Recommender That Chose the Right Metric
A Hack Week project became 20% of Spotify's streaming volume. The machine learning is the least interesting part — what made it work was a constraint most personalization products refuse to accept.