Formidable?

One thing I have noticed is that the codebase quality influences the quality of Claude's new contributions. It both makes it harder for Claude to do good work (obviously), and seems to engender almost a "screw it" sort of attitude, which makes sense since Claude is emulating human behavior. Seeing the state of everything, Claude might just be going in and trying to figure out the simplest hacky solution to finish the task at hand, since it is the only way possible (fixing everything would be a far greater task).

Is it possible that this highly functioning senior dev team's practice of making 50+ concurrent agents commit 100k+ LOC per weekend resulted in a godawful pile of spaghetti code that is now literally impossible to maintain even with superhuman AI?

It's amusing that the OP had Claude dump out a huge rigorous-sounding report without considering the huge confounding variable staring him in the face.

Obviously it's going to be more productive for a manufacturer to do a years-long curing test on 100 likely candidates instead of 100 random mixes. They obviously already screen candidates through traditional methods, but if this AI technique improves accuracy, all the better.

He seems to be trying to take almost a "venture studio" approach by throwing shit at the wall, but the problem with these things is always that the "internal startups" are "founded" by people who don't have enough incentive or control over their product to perform as well as an actual startup, and are distracted by internal politics. And frankly, it may also be that the really good founders will just do their own startup vs working on a quasi-startup inside a large org so there's some selection bias as well.

This only works if there are enough people betting on the other side. It's not some kind of magical money multiplication machine. As more stuff like the one-day Iran bet or the 64:56 minute press conference happen, people will avoid taking bets on highly specific outcomes.

On more reasonable bets like "war with Iran in the next 6 months", if there is some kind of shadowy cabal putting billions of dollars on the yes side, they just aren't going to have much upside if there isn't the same amount on the no side.

LLMs mostly do useful work by writing stories about AI assistants who issue various commands and reply to a user's prompts. These do work, but they are fundamentally like a screenplay that the LLM is continuing.

An "agent" is a great abstraction since the LLM is used to continuing stories about characters going through narrative arcs. The type of work that would be assigned to a particular agent can also keep its context clean and distraction-free.

So parallelism could be useful even if everything is completely sequential to study how these separate characters and narrative arcs intersect in ways that are similar to real characters acting independently and simultaneously, which is what LLMs are good at writing about.

Seems like the important thing would be to avoid getting caught up on actual "wall time" parallelism

An LLM running one query at a time can already generate a huge amount of text in a few hours, and drain your bank account too.

A "different agent" is just different context supplied in the query to the LLM. There is nothing more than that. Maybe some of them use a different model, but again, this is just a setting in OpenRouter or whatever.

Agent parallelism just doesn't seem necessary and makes everything harder. Not an expert though, tell me where I'm wrong.