The problem
Memberships touch almost every part of our platform: checkout, payments, renewals, refunds, emails, scheduling and notifications. Our normal estimate was three months, or four to five with proper testing.
AI coding tools can produce a demo in an afternoon. The question was whether they could build something this connected to production standard, safely, on a live platform our customers rely on.
What we did
Documents before code
We spent three to four days writing the spec, the decisions, our engineering and design standards, and a phased build plan. The agent worked from those documents, not from a prompt.
The same loop for every phase
Plan, approve, build, test, review, fix, commit, hand off. A person approved each plan before any code was written, and reviewed every high-risk change before it was committed.
Our own standards as the bar
Every change was checked against the same rules we hold our engineers to: security, testing, design and documentation.
The principle
Write the standards first, then hold the AI to them like any engineer.
The results
The build took about 14 hours of AI time over two days, after three to four days of planning: 241 small, reviewed commits and 735 automated tests.
Reviewing the work against our standards also surfaced problems in our existing platform, which we then fixed. That review was worth doing on its own.
What this means for your company
The speed comes from the method, not the tool. Most teams try AI coding without written standards or a review loop, and get fragile code fast. A Guided Build installs the method we used here, so your team can repeat it on its own.
Figures from our own build log.