As I point out in my white paper on AI-augmented software engineering, the median engineering team spends 20-40% of their effort on rework. That is, every week of work generates another 1-2 days of fixes.
Software product teams commonly cut corners in the name of “velocity”, deliberately adding future rework to get results now. This doesn’t work. You’re paying the 20-40% tax on the work you’re choosing to do now (nothing about “cutting corners” makes the immediate work higher quality); you’re paying for the work you defer, and for the inevitable 20-40% of rework that deferred work also generates.
That’s a lot more extra work than it seems upfront, so what happens when you get there? You want to get to more value quickly, so you deprioritise the fix work in favour of new features. The rework piles up. The team slows down. Customers no longer feel empowered, because the software doesn’t make them awesome, it makes them practised at working around your bugs. Eventually a latent defect—maybe a security issue, maybe a production outage—becomes an existential crisis.
Good teams can get down to 10% rework, and they do it by focusing on quality first. Bugs don’t get added to “the backlog”; they get fixed, and the team learned from them. Sure, make some big experiments, but don’t force your customers to deal with them in production.
