Blog
What we learned building production AI systems. No theory — just what happened, what broke, and what we'd do differently.
Not a tutorial. Not a review. Just the honest reality of using Claude Code as your primary dev tool every day for over a year.
Six tools that turn Claude Code from a coding assistant into a full operating system. Adversarial reviews, autonomous experiments, persistent memory, and 200+ app integrations.
A persistent Ubuntu machine you can talk to in plain English. The product is interesting. The pitch is dishonest. Both can be true.
Open CLI for browser automation, Claude Managed Agents for persistent sandboxes, and Ultrareview for code quality. This is the stack that's actually working in production.
I built a Stripe app the traditional way — 28-day recovery sequences, ML-optimized retries, a full dashboard. Now I'm watching AI agents make the whole concept of 'apps' obsolete.
I built an AI-powered payment recovery app and submitted it to Stripe's App Marketplace. Here's what their review team flagged, what I had to fix, and what I wish someone had told me before I started.