It's not complicated at all. You don't "manage agents". You just type your prompt into an terminal application that can update files, read your docs and run your tests.
As with every new tech there's a hell of a lot of noise (plugins, skills, hooks, MCP, LSP - to quote Kaparthy) but most of it can just be disregarded. No one is "behind" - it's all very easy to use.
Easy to use, hard to master. Or: low skill floor, high skill ceiling. My output wouldn't be nearly as good without subagents and skills, and MCPs are somewhat required if you deploy tool using agents at scale.
It's like saying all you need is notepad to develop. It's not wrong, but.. you know.
It’s not hard to master. It’s not a skill to be learned —- it’s a tool that comes with a manual. You read the manual and now you can use the tool. Most people never will read the manual which is what gives the false impression that there’s something “to master” here. It’s like saying vím is harder to use than notepad. Not if you read the entire manual first.
I'm not sure how you define skill acquisition, it's reading documentation and doing the skill, yes? The AI landscape shifts rather quickly still, and a new LLM + harness has a different set of functionality, but more importantly different fuzzy failure cases Things a model is particular good at, things that work better if you combine certain systems. All of it is documented, but also fast moving and new things are discovered frequently. In comparison, Vim has been around for decades.
And vum is absolutely harder to use than notepad. Otherwise it's like saying that rocket science isn't hard because you just have to read the documentation to know how to engineer a rocket.
This is the start of the end of generative AI music.
Without the entire music catalog, they are just not going to be able get the diversity and quality of outputs they once had, and the labels are always going to have their hands out wanting more.
Udio and Suno have demonstrated that okay-quality new music can be created by training on vast collections of pirated music, and, as expected, the music labels shut them down.
I find it useful. A nice little tool in the toolkit: saves a bunch of typing, helps to over come inertia, helps me find things in unfamiliar parts of the codebase, amongst other things.
But for it to be useful, you have to already know what you're doing. You need to tell it where to look. Review what it does carefully. Also, sometimes I find particular hairy bits of code need to be written completely by hand, so I can fully internalise the problem. Only once I've internalised hard parts of codebase can I effectively guide CC. Plus there's so many other things in my day-to-day where next token predictors are just not useful.
In short, its useful but no one's losing a job because it exists. Also, the idea of having non-experts manage software systems at any moderate and above level of complexity is still laughable.
I don't think the concern is that non-experts would manage large software systems, but that experts would use it to manage larger software systems on their own before needing to hire additional devs, and in that way reduce the number of available roles. I.e. it increases the "pain threshold" before I would say to myself "it's worth the hassle to hire and onboard another dev to help with this".
As with every new tech there's a hell of a lot of noise (plugins, skills, hooks, MCP, LSP - to quote Kaparthy) but most of it can just be disregarded. No one is "behind" - it's all very easy to use.
reply