You still have to know enough to tell the AI what you want in more detail than we generally get from project managers. And if you have an existing codebase, as you do 99% of the time, you need to put all that in context so the AI can "understand" it well enough to review it before you can have it alter the code per specs.
I don't think the day is far off when AI can do most of the work, but right now we're seeing a lot of demos of simple apps. And it will regenerate the app to add a feature. That will not cut it with enterprise-level systems. Most of the apps I've worked on over the last few decades have run on multiple servers, and many were even spread across multiple "projects". The basic config for one spread over was 24 servers. That one had fully automated CI/CD, but you would be shocked at how many projects I've been on without any automated testing before I got there. And little documentation. If you do not have that with the current codebase, how are you going to have AI take it over? Not to mention projects with a change control board that must sign off on every commit to the main branch. It's going to be a while before they approve AI-rendered changes without a developer explaining the change to them. Then there is the whole private repo issue. Those projects will not have code generated by an external AI. But again, self-hosted AI is rapidly becoming more common. Just not that common yet.
No comments:
Post a Comment