You can still do both. Its just all of the grunt work is no longer grunt work, and all the tech debt you've been putting off is no longer an issue, and all of the ideas you've been meaning to try out but don't have the time can suddenly be explored in an afternoon, and so on and so forth.

For new features, by all means code it by hand. Maybe that is best! But a codebase is much more than new features. Invaluable tool.

If you don't see how that fits into "group 2" in GP's comment even though it wasn't explicitly called out, then we may have identifed why you don't find agentic coding to be enjoyable.

That's true, I really don't enjoy expressing my ideas in natural language because I already see them as code in my head. So, what's the point of translating the code in my head to human language to produce the code? I can just write the code, lol. Vim shortcuts + two-hands typing and you are fast enough.

Have you considered writing code to the agent? It doesn’t have to be natural language.

When you start to integrate MCPs it feels especially like programming in a new language. You basically end up adding commands to your tool.

I already know like five of them, it's more than enough to write anything my weird brain comes up with :D. I'm fine with the setup I have now, if I ever would have to use agentic coding because my employer would demand doing things faster, for the pure sake of doing things faster, I will just quit. If everyone started demanding that, I will just say "goodbye" to my IT career for good because that wouldn't be fun anymore; yet another white collar job.

It's only about me having fun from coding. The low-level stuff is the most fun part. I don't know why is it so hard to swallow pill.

I think you might have identified where it would help you the most then.

So if your preference is to avoid high level stuff then genetic coding may be useful for, as an example, updating your suite of tests after a major refactoring. Or if you have no tests because that may not be your style, then generating some robust test scripts might be a good use case.

Or maybe you create one good test but you need to expand it with a lot of various alternative edge cases. Agents can generate the edge cases for you based on the test you already wrote.

Or writing high level plumbing like checking if the scopes are all correct still in your dependency injection scaffolding. Etc

It sounds like you’re highly defensive of the part of your job you like, which is commendable of course, but there may be opportunity to improve the parts you don’t like.