Getting six product people to change how they work

Nobody asked for this. I proposed it, and I kept pushing the CTO until I had budget and room to build it. That is the part of this case I would point to first: the work only existed because I sold it internally against a plan that did not include it.

The problem was that our work was disconnected between phases. Research lived in one place, prioritisation frames were rebuilt from scratch every time, and none of the company context sat anywhere a tool could reach. Every stage had its own bottleneck and nothing carried over.

What I built is a shared Claude Code environment with three layers. On top, workflows: slash commands and skills that produce the same correct structure every time, for PRDs, business cases, prioritisation, meeting notes. In the middle, four specialist sub-agents, an engineer for feasibility, a data analyst, a user researcher and an executive lens, so a PM can ask the CTO's question early rather than in the review where it costs a week. At the bottom, company knowledge: platforms, the order state cycle, personas, glossary, Jira conventions.

The bottom layer is the one the whole thing rests on. It loads automatically underneath every command above it, so nobody has to remember to explain what MIP is or which squad owns Odoo. Output quality stops depending on how well each person happens to prompt.

It is distributed over Git: one shared base, plus a personal folder for each person so nobody steps on anyone else. I ran the training. I was also the first user, then taught one PM and the Product Designer, and it spread from there. Two full-stack tech leads were rolling out spec-driven development and Cursor on the engineering side at the same time, so the two halves were built to meet.