[JTbane]

No AI company is addressing the elephant in the room that you need someone experienced constantly monitoring any agentic workflows. This means that the cost savings of agents are a myth.

My company actually did an internal study of agent usage for coding and found it only improved productivity by 10-20%, basically on the same level as good code templates or an autocomplete.

[bentcorner]

I maintain a part of my team's CD process and I've observed a 30% increase in PR velocity since we started adopting agentic tools but it was a "one-off" increase (as-in, it hasn't continued to increase beyond that since about a half-year ago).

I'm guessing though that there are other improvements in code quality and feature velocity. I've noticed personally that AI is really good at catching smaller things that are easy to miss (e.g., if you ask it to rename fooTheBars it also updates all the relevant comments or enums that you might have missed).

[disgruntledphd2]

> My company actually did an internal study of agent usage for coding and found it only improved productivity by 10-20%, basically on the same level as good code templates or an autocomplete.

That's still a pretty good outcome. 20% more output across a company is huge when you think about it. Definitely not going to change the world completely though.

> No AI company is addressing the elephant in the room that you need someone experienced constantly monitoring any agentic workflows. This means that the cost savings of agents are a myth.

I mean, it depends on the agentic workflow. Like for production code, definitely. For document and claim review, you probably need a targeted sample on a daily basis but you get massive gains.

[ACCount37]

Less and less true with every new generation of AI systems.

AI gets better and better at operating self-supervised, and the amount of skill needed to supervise an AI in a useful fashion only ever goes up.

[bigstrat2003]

That simply isn't true. LLMs are completely incapable of operating without supervision, same as they were 3 years ago.

[ACCount37]

The scope and scale of the tasks LLMs can be trusted to do without supervision has increased massively in the meanwhile.

Of course, it will never be enough. The goalposts will move until we run out of them.

[array_key_first]

I mean, what's the goal here? AI does everything and then... and then what? We all die? Most of us die except a select few who rule the world (which, by the way, would not include you or me)?

The reason we won't let AI run rampant is because it's anti-human. Even if AI can do everything, why would we allow it to? We don't have to do that, that's very much a choice.

We're at the point where we have humans advocating not only against other humans, but against themselves. If people are going to be out here arguing for self-destruction on the order of suicide, they might as well save us all the trouble and put a gun in their mouths. At least that way, humanity might have a few less pathetic losers who put computer programs before themselves.

Maybe that's harsh, but that's how I see it.

[ACCount37]

The alternative is to let humans do everything. Think about what that entails.

I have seen enough human incompetence that AI incompetence is beginning to look quite favorable. AI performance, at least, goes up generation to generation. If you've ever faced minimum wage workers, you know that they are, at best, consistently the same level of awful.

[array_key_first]

How is that the alternative? That's not even the reality 100 years ago, let alone today.

[zelphirkalt]

If we move the goalposts along, how can we run out of them? No, I find your argument flawed, sir.

But if we move us faster than the goalposts, we can outrun them.