
The 60/20/20 building rule

Why PMs are suddenly valuable again | Tulsee Doshi (Head of Product, Gemini)
Topic 11 of 18
Published Nov 8, 2026 · Clip 33:16–36:43 (3:27)
Tulsi explains her 60/20/20 rule for how far ahead of model capabilities to build. Since models constantly improve, traditional 'build an MVP that works' wisdom is incomplete: build so ~60% of what you build works now, ~20% works well enough with a clear path where models will get there soon, and ~20% targets capabilities where you see glimmers of light and which would create a magical experience if cracked. This balances shipping something great today while setting up to grow as models grow. The host compares it to a portfolio of product bets, and Tulsi adds that products should stay agile—rigid systems built around old model capabilities get outgrown, so retest regularly rather than force-fitting new models into old structures.