The biggest change I have felt in the last year or two is, *the amount of pull requests has increased abnormally.
I was an engineering manager. I was looking after a team of more than a dozen people, More than half of them were contractors (secondary engineers).
---
They operate at night and on weekends after their day job. And from Friday night to Sunday night, Massive numbers of pulls go up simultaneously.
One contractor may work on several repositories at the same time, By the end of the week, they are in a state of "where do I start?". Dozens of pull requests piling up over the weekend alone are commonplace.
---
Just because a product is a microservice configuration, Even the merging process is not micro.
In practice, the business logic is complex and spans across each service, If merged without understanding the context, The integrity of the whole system is instantly destroyed.
At first glance, productivity appears to be increasing. But in the aggregate, the product becomes more unstable.
---
It is difficult to rest in this situation. If you leave Saturday and Sunday's pulls unchecked, Monday, something might be broken somewhere.
After all, managers and developers, "structure that cannot rest".
---
Superficially, this might be described as 'productivity gains'. But essentially, there is a growing 'loss of context'. Productivity is, Not about writing code faster, *depends on the ability of people who understand the context to make the right decisions and merge correctly.
---
AI can assist in this area in some areas, though, the limitations are also clear. Particularly in UX and front-end, where 'eyes' and 'ears' are needed, AI cannot yet complement human senses.
AI can proceed with just the back-end logic, though, UI and experience design cannot.
The more AI is used, the more productivity is said to fall, *The more AI is used, the less productivity is achieved, and it is easy to fall into the "reverse productivity trap ".
---
This is a problem that becomes apparent when you get beyond a development organisation of a dozen or so people. I'm not talking about things like "01 prototypes in Claude Code." *It's a reality that emerges when you get to the stage of maintaining a product that works on a production scale.
Both AI and humans are already working to their limits. Still, there is nowhere to guarantee that products will not break down.
