State of AI4SDLC: How Software Development Changes as Code Becomes Easier to Write

The recording of my September 25 DotNext keynote is out. There are also written notes and slides, so you can pick the format you prefer. I shared the main points back in September. Now you can watch the full talk, including stories from practice. The central idea comes from Chad Fowler's book “Regenerative Software,” which I briefly covered. In short, the question is this: as the implementation becomes easier to replace, what knowledge about the system needs to outlive its code?

In the talk I covered:
– Why speeding up code writing alone doesn't guarantee faster delivery. The bottleneck can move to task definition, review, and acceptance.
– What needs to survive outside the implementation: intent, architectural boundaries, behavioral checks, and the reasons behind decisions.
– Why migration doesn't end when the new system launches. You still need to switch off the old one and remove the complexity it accumulated.
– How an engineer's job changes when an agent writes the code and a person remains responsible for the result and the system's future.

I use examples from migrations I've worked on, requirements that weren't obvious, and decisions whose rationale can easily disappear in the next rewrite.

Have you had a new system up and running while the old one somehow stayed on? Tell me in the comments. I'd love to hear what was keeping it alive :)

#AI4SDLC #AI #Engineering #Architecture #Software