I read the line and thought it was AI, and then saw the username and figured it wouldn't be. I find my writing being influenced by LLM patterns as well, because that's basically all the writing that exists these days. I don't like how it erodes my own voice because I keep reading the writings of this one single author all day, every day.

FWIW, I thought it was human written. Pangram also said that it's 100% human written.

More likely, they're all wrong on this accusation. A bit concerning to see how many of them are throwing out a false positive like this. Makes me wonder how many of them are falling for actual AI traps in the real world. :)

I am actually with you here. Such a system probably needs to be battle tested in a specific domain of business to see where the deficiencies at this stage are.

Problem is, no matter how many sub-agents you break the LLM into it's still stochastic parrots all the way down. I've seen no evidence that compartmentalised harnesses can reach solutions that a monolithic harnesses cannot, so I'm thinking the compartmentalisation is just an excuse to throw more tokens into the token furnace - it doesn't unlock a step change in capability. If there is evidence to the contrary, I'd be interested in seeing it.

There have been some successful attempts to do science like this (https://github.com/AstroPilot-AI/Denario). Denario generated 1k+ "solutions" on its way to winning the FAIR Universe challenge, and they all had to be fed through a fitness function of some kind to assess them. It also got stuck on a local maximum and had to be kicked in the ass by a human to get it off that.

When writing software, that fitness function is relatively closed (when I run it, does it do what I want?). With something as open-ended as managing a business... might as well roll dice, IMO.

My belief is that 20's LLMs have a lot in common with 90's GAs: the more clearly we can define "success" for a given task (the fitness function) the more successful they can be. Open-ended problems are somewhat beyond them right now, and, I suspect, forever.

(The comment that's currently at the top of the discussion, about making an "AI Boss" that remembered everything about the company, and fed the user three tasks a day? That's not an AI Boss, that's an AI Assistant with different framing.)

Well, you mentioned Gas Town — that deserves an immediate reflexive downvote :P

I actually upvoted your post. I'm not sure I entirely agree with it, but I was interested to see more discussion about it, and that felt worthy of an upvote.

As for the reaction against the idea, I see it from two angles. Is the standard corporate hierarchy / architecture really the optimal organization principle for AI? Or is that just a skeuomorphism to make this project seem more serious than it really is? Maybe Yegge did us a favor by mapping to Mayors & Polecats, forcing people to question if it really is the correct organizational hierarchy?

The other angle is heresy here - HN is full of employees now. There is little incentive for employees to be interested in what a C-suite does, and we see that with the comments about CEOs needing to be six foot and walk impressively. Employees have no interest in AI organization principles that optimize them away and highlights their obsolescence. I am surprised there hasn't been more discussion about how this specific project optimizes away the C-suite humans instead - I would have thought that could have had some appeal to employees, at least until those people realized it makes employees a swarm that works for AI leadership.

As for AI organization, the organizational mapping needs to be productive, not affectation or cargo-culting. That's not peripheral — that's load-bearing. Especially in a corporation model, you can find the organization layers only create busywork with lots of unnecessary middle managers. You don't want your AI C-suite swarm burning tokens on endless meetings about what should be in the upcoming meeting... especially not at Fable API pricing.