Almost everything. What can be done in an hour? The biggest issue for me is I have no idea how to make AI work across multiple services or dozens of packages. Maybe the organization and software needs to be built from the ground up with AI in mind?

For example, I’m rebuilding a legacy backend application and need to do a lot of reverse engineering. There are a dozen upstream and downstream services and nobody in the world knows 100% of what they do. AI doesn’t know how to look across wikis, slack channels, or send test requests and dig through logs but the majority of the work is this because nobody knows the requirements. also a lot of the “code” is not actually code but several layers of auto-generated crap based only on API models. How can I point an AI at the project, say “here’s 20 packages involved with a 3-service call chain that’s only 2/3 documented” and get something useful?

The code is pretty much always the easiest part of my job.

You can try to tell me that this is actually a symptom of a deeper problem or organization rot or bad design choices, and I agree, but that’s out of my control and my main job is to work around this crap and still deliver. It was like this for years before 90% current employees were hired.

To summarize, I work in micro-service hell and I don’t know how to make AI useful at all for the slow parts