cjroth.com
|
ksl
|
|
Chris Roth compiled practices from Linear, Cursor, Vercel, Stripe, and Resend to map what separates teams that compound AI gains from those drowning in AI-generated bugs. The data is specific: senior engineers realize nearly five times the productivity gains of juniors, median PR size grew 33% in 2025, and top AI-native startups hit $3.48 million revenue per engineer versus $610K at traditional SaaS companies. Cursor reached $500 million ARR faster than any SaaS product in history while shipping from a single monolith every two to four weeks. The throughline across all these teams isn’t tool selection – it’s spec-driven development, stacked PRs, dissolved design-engineering handoffs, and AGENTS.md files that give AI agents enough architectural context to stay useful. Teams adopting AI without that underlying discipline are finding it just makes their existing problems louder.
