Case 02
From prototype to pull request, with AI at every step
BigBuy · Product Manager · 2026
A delivery chain where AI does work at every step and a human signs every handover.
- 2 sessions
- To produce the full Mailbox PRD from its prototype
- Half
- The time to get from a design to a specification
- 1 sprint
- Dedicated to AI-assisted delivery, in the company's own Jira
- 2 MCPs
- Figma Make and Atlassian, in a shipping process rather than a demo
Turning a design into a specification used to be a person writing for hours, and then explaining it again in refinement. We replaced that with a chain where every step produces the input for the next one and a human signs off on every handover.
Product prototypes in Figma Make. The specification comes out of the prototype through the Figma Make MCP: a complete PRD with every function and feature, explicit MVP scope and explicit non-scope, not a summary engineering has to interpret. Engineering refines technically against that document, and any trade-off that surfaces comes back to product to be decided. PMs create the Jira tasks through the Atlassian MCP. Developers point Cursor at those tasks, give it instructions, and review the pull requests it produces.
We ran a Jira sprint dedicated to working this way. That is the part that is hardest to argue with: it exists in the company's Jira with tasks, commits and reviewed pull requests behind it, not in a story I tell about it.
The Mailbox MVP went through this loop end to end. Its PRD, every function and feature of it, came out of the Figma Make prototype in two working sessions. The old way meant writing all of it down by hand first. Across the team, getting from a design to a specification now takes about half the time it used to.