Byatnal’s Extend hit $1M+ ARR with 3 engineers on the most boring possible stack: React, Node, TypeScript, Postgres, Redis. Core stack unchanged through 2.5 years of growth. The counterintuitive insight: this simplicity is a competitive advantage specifically because AI coding assistants work better with standard, well-documented technologies.
The monorepo decision compounds the advantage — all context in one place makes Claude and Cursor “dramatically more useful.” Byatnal’s one regret: not putting the docs repo in the monorepo too. This directly supports Context is the product, not the model from an engineering angle: giving the AI assistant more coherent context (monorepo) produces better output than giving it a smarter model.
The selective additions tell the story too: Go and Rust only for CPU-bound work (file conversion, Excel parsing), AWS migration for enterprise compliance, WorkOS for enterprise auth. The pattern is infrastructure-over-middleware applied to startups: don’t build your own auth or queue system, pay for proven infrastructure and spend engineering time on your actual product. At the database layer, Lakebases decouple compute from storage — databases become elastic infrastructure represents the next evolution of this philosophy — compute and storage separate entirely, so the database becomes elastic infrastructure that scales to zero rather than a fixed cost you architect around.