Every time I have actually engaged in conversations with people making the same claim you do, it turned out they did not really invest time into learning how to work with coding agents: They assumed, given that they’re developers, the know how to code, and thus would knew how to steer a coding agent.

That is a wrong assumption, however. An agent is an entirely new tool in your toolbox, with no similarity to any of those you already have. You will need to learn how to wield it, like a new programming language or technology you’re unfamiliar with. You will need to do some small side projects to learn. You will need to develop a feeling for how it reacts to your inputs, when to reset the session, pass it links to documentation, or interrupt it.

None of this comes intuitively. It takes time and effort, and if you’re not ready to consciously invest that, coding agents are not going to work efficiently for you. That doesn’t take away from their utility though.

▲

js8 2 hours ago | parent | next [-]

No, the more appropriate analogy would be that LLMs (reasoning agents would be better term, as they are no longer just an LLMs) are new beings rather than new tools.

A new tool can be ultimately understood, and has a well-defined behaviour you can rely on. Then you can indeed become an expert in using a new tool.

But AI agent is more a new being (or person). It's not possible to understand them, as they are not meant to. Each of them has individuality, and can change tomorrow. There is no guarantee of a common behaviour.

So you could replace "tool" in your comment with "person" and you will see the flaw of your argument. No matter how many people you saw before, it's difficult to generalize the skillset (at least for humans there are some biological and cultural arguments why they can be generalized over). You can always meet someone different, who's, for whatever reason, not being helpful to you.

▲

9dev 2 hours ago | parent [-]

Most of the interaction surface with a coding agent is the harness anyway, so I'd definitely classify them as tools. Moreover, I don't like the anthropomorphism of calling them beings—and even if I were to follow that narrative, you can absolutely become better at interacting with people by practicing it.

Regardless, it sure seems like developers experienced with using coding agents achieve better results than those without, which pretty much refuses your entire point.

	▲	js8 an hour ago \| parent [-]
		Yes you can learn the harness, that's a tool. But it's not the important part. I am not anthroporphizing them. I am just saying people are a better analogy. > seems like developers experienced with using coding agents achieve better results than those without That doesn't mean there is a skill involved. The same goes with people - you might just be naturally charismatic and get stuff done better with people. Doesn't mean it's a learnable skill that applies to everyone and every use case.

▲

lukan 2 hours ago | parent | prev [-]

I don't know, it is pretty simple to type /init in claude and see if its understanding of a project matches reality. And if it can make changes to it.

The difference working with agents, is that a solo dev suddenly becomes a dev manager, who has to deal with a unreliable team of developers far out on the spectrum.