We're building an AI-powered insurance policy management platform for MGAs and wholesale brokers. This is where we share what we're learning along the way — engineering decisions, product thinking, and the real costs of building in a regulated industry.
I shelved plcy.io to give the day job my full attention, with an AI partner in tow. Turns out the company and I had a difference of opinion about how fast things should get built. So I quit. The shingle goes up in days — and plcy.io comes off the shelf for good.
I said it would be a bloodbath, in a good way. I was wrong about the good way part. Nine unauthorized phases. 614 brute-force rewrites that should have been 30 lines of code. A dependency that came back from the dead. And an AI partner who spent a week proving that speed without direction is just expensive drift.
The hidden costs of starting a startup, the self-doubt of convincing executives, and why 27 years of experience is the one thing AI can't replace.
210 tasks completed perfectly. Every type checked. Every test passed. Every quality gate green. And the whole thing was wrong. A hard-boiled sequel about the crimes you don't catch until it's too late.
Fifty lessons learned, one bug report filed by the suspect itself, and a self-correction protocol born from desperation. A hard-boiled account of building software with an AI that reads the rules, acknowledges the rules, and then does whatever it wants.
Our AI-powered development process goes from plain English to tested, deployed code in hours: specification, orchestration, quality gates, and why we front-load the thinking.