Building Real Software, Part-Time, With AI

Starting from a blank repository in November 2025, the platform was built using a combination of large language model tools for implementation, at fewer than fifteen hours a week. AI generated nearly all implementation code, including corrections and test scaffolding, while the judgment calls — architecture, security requirements, acceptance criteria, and what not to build — stayed with the developer.

A disciplined operational structure backed the build: three environments (development, staging, production), each with its own database branch and secrets, a CI/CD pipeline with an automated quality gate, and a human-approved production gate that no automated system can bypass. Ahead of launch, a structured internal security assessment simulated an external assessor's approach, surfacing five findings — including a Broken Object Level Authorization (BOLA) vulnerability — all resolved with regression tests before go-live.

The result: a launch-ready platform with roughly 80,000 lines of code, 189 API handlers, and a guarded three-environment deployment pipeline — built in approximately 450–500 hours over eight months. A conventional solo build was estimated at 2,300–3,100 hours, putting the apparent compression in the range of 5x–6x, achieved by one experienced practitioner working part-time.

The full account is a candid one, findings and all: it documents what AI accelerated, where it fell short, and the honest limitations that remain at launch — including the specific vulnerabilities found and fixed before go-live.