The Problem
On December 28th, I typed "add a goal tracking agent" into Claude Code. Three hours later, a complete intentions agent—backend, frontend, voice integration, and website update—was in production. One commit. Nearly 3,000 lines. Twenty files.
Three days later, I shipped a feedback and memory system: users can thumbs-up agent responses, the system learns from patterns, and agents remember preferences across sessions. Six commits over three days. About 4,500 lines.
On January 1st, a complete daily summary email system. Celery task queue, distributed scheduling, timezone handling, Kubernetes manifests. One commit. Production by evening.
This isn't about typing faster. It's about building a codebase around how AI coding agents work.
Traditional development is slow because coordination is slow. A new feature touches the backend, the frontend, maybe infrastructure. That's three teams, three repositories, three deployment pipelines. A backend engineer opens a PR, waits for review. The frontend engineer can't start until the API exists. DevOps sets up the infrastructure last. Calendar time stretches from days to weeks.
Context switching kills productivity. The backend engineer finishes and moves on. When frontend integration reveals a problem, they context-switch back. The mental model is gone. They reload it. Time passes.
We call it an agent factory: a monorepo designed for AI-assisted development—optimized for how tools like Claude Code understand, navigate, and modify code. The development environment is as much a product as the agents it produces.
