60 engineers heading to 150. Everything from 30 is straining. What broke first crossing 50 to 200?
What broke first crossing 50 to 200?
For us alignment broke first. Two ML pods spent a full quarter building the same feature store, neither knowing about the other until a joint demo made it painfully obvious. Both teams were optimizing locally and hitting their own metrics, which is exactly why nobody flagged the duplication. Past 50 to 75, that pattern repeats everywhere unless ownership is explicit. Implicit ownership does not survive the cliff. What fixed it for us was a boring but effective experiment: one named team per pipeline, written down, with decision rights. No shared maybes. Michelle, curious where you are on this, is ownership written down anywhere yet, or still mostly tribal knowledge?
I’d push back slightly on alignment breaking first. The org chart is downstream of the architecture. Two teams collide because they share one deploy queue and one serving path, so every reorg is just tax on coupled systems. Redraw the seams before adding managers, because what worked at 50 won’t work at 200. What is still shared that every team touches?