Engineering leadership · Book notes
Project to Product and the work a feature roadmap hides
Flow distribution tells you what cheap features crowd out: defects, risk and debt.
You’re looking at your team’s end-of-quarter done column. All the cards are features. You advanced your roadmap, you successfully demoed. But is this a product team? As Mik Kersten describes in Project to Product, look at what’s not in your column. The Flow Framework describes all software work as four flow items: customer-pulled features, defects, risks (security and compliance work), and debts. If your column only has features, it means you didn’t fix any defects, you didn’t deliver any risk work, you didn’t pay down any debts. Or you did, but you’re not counting it anywhere.
I think the headline idea in this book (projects end, products don’t) has largely aged into our vocabulary. We won it as a way of talking, but lost it as a way of funding teams. I would keep the flow distribution as a measure to determine if that shift really happened. The four items all share from the same fixed capacity, so a larger share of features means a smaller share for the other three. Your roadmap only lists features.
Kersten warns of this in his essay how to budget for technical debt. As more new functionality gets delivered, less time is available for fixing defects and doing security work. A shrinking share of work on technical debt is a leading indicator for defects and delays that haven’t yet become visible. The question for me is: when you are in a planning meeting, who argues for the share of technical debt when features are being demoed? Kersten says you need to reserve time and money for technical debt every financial year. If you don’t, the debt will remain invisible to the business, just like it did for Nokia with Symbian.
The cheap features of AI coding agents make this problem even more acute. If an agent can build a feature in an afternoon, the share of features will grow unprompted, while risk and debt only get their share if deliberately protected. The trade-off gets worse before it gets better. If agents can also fix defects and pay down debt, does the trade-off go away, or does the fixed capacity shift to people reviewing the agent work?