| Target Entities | AI Factory, Software Delivery Contract, Change-Size Limits, Release Control, Verification Rules, Procurement Software, Spend Limits, Human Deployment Approval |
|---|---|
| Core Value | Written change-size limits, repeatable verification and human deployment decisions, with purchasing authority kept separate from code acceptance |
| Tech Stack | AI Factory delivery contract, Automated test harness, CI, Pre-push size check, Exact-commit release builds, Simulated vendor checkout |
Software & AI
How an internal software-buying build went from 69% to 92% of changes within agreed size limits
An internal software-buying platform reached live preview, with its buying journey validated against a simulated vendor. Payment-provider and pricing decisions were still open. On day 54 of the 101-day build, we put the work under a written AI Factory delivery contract. Changes within agreed size limits then rose from 69% to 92%.
changes within agreed size limits, before vs after the day-54 delivery contract
commits over the 101-day build
internal product, buying journey validated against a simulated vendor
Challenge
Build a software-buying platform where a wrong change can affect payment behavior.
What we built
A written delivery contract with change-size limits, then automated checks and exact-commit releases.
Result
92% of changes within agreed size limits after the contract, up from 69%, with purchasing authority still separate from code acceptance.
The challenge
A software-buying platform has to connect purchasing requests with signed authority and hard spend limits. A wrong change can affect payment behavior, so development also needs clear checks and clear release ownership. Early work moved quickly, but automated testing and review rules were incomplete. Review found tests that passed even when the behavior they were meant to check had not run.
What we built
- Introduced a written AI Factory delivery contract on day 54 of the 101-day build, with explicit change-size limits. Larger changes needed a person to read them before acceptance.
- Added automated checks and a pre-push size check. Changes to verification rules needed separate review, and behavior fixes needed a check that failed before the correction.
- Built release images from an exact committed version, reran the checks on it and left the deployment decision with a person. The contract granted no purchasing authority: a real-money run required a deliberate live-spend switch, credentials and a run-specific ceiling.
How the data flows
Change controls
Release and spend
The outcome
Changes within agreed size limits rose from 69% before the contract to 92% from its introduction onward. Checkout verification used a simulated vendor. Review also caught a case where a canceled subscription could still be charged, and the defect was reproduced and fixed. The product remained in live preview with payment-provider and pricing decisions open: internal development progress, not a completed customer rollout.
Systems & technologies
Before vs after
Get in touch
Building software where a wrong change can affect payment behavior?
We can run your build under a written AI Factory delivery contract, where a person reads larger changes and makes the deployment decision.
Email us directly
sales@3alica.com