Would you ever trust an AI agent running your business? As hilarious as this small experiment is, is there ever a point where you can trust it to run something long term? It might make good decisions for a day, month or a year and then one day decide to trash your whole business.

Bye bye, B2B. Say hello to Ai2Ai.

No humans at all. Just Ai consuming other Ai in an "ouroboros" fashion.

The "April Fools" incident is VERY concerning. It would be akin to your boss having a psychotic break with reality one day and then resuming work the next. They also make a very interesting and scary point:

> ...in a world where larger fractions of economic activity are autonomously managed by AI agents, odd scenarios like this could have cascading effects—especially if multiple agents based on similar underlying models tend to go wrong for similar reasons.

This is a pretty large understatement. Imagine a business that is franchised across the country with each "franchisee" being a copy of the same model, which all freak out on the same day, accuse the customers of secretly working for the CIA and deciding to stop selling hot dogs at a profit and instead sell hand grenades at a loss. Now imagine 50 other chains having similar issues while AI law enforcement analysts dispatch real cops with real guns to the poor employees caught in the middle schlepping explosives from the UPS store to a stand in the mall.

I think we were expecting SkyNet but in reality the post-AI economy may just be really chaotic. If you thought profit-maximizing capitalist entrepreneurs were corrosive to the social fabric, wait until there are 10^10 more of them (unlike traditional meat-based entrepreneurs, there's no upper limit and there can easily be more of them than there are real people) and they not-infrequently act like they're in late stage amphetamine psychosis while still controlling your paycheck, your bank, your local police department, the military, and whatever is left that passes for the news media.

Deeper, even if they get this to work with minimal amounts of of synthetic schizophrenia, do we really want a future where we all mainly work schlepping things back and forth at the orders of disembodied voices whose reasoning we can't understand?

It would be cool to get a follow up on how long it's been since this write up and how well it's been doing since they revised the prompts and tools. Anyone know someone from Andover Labs?

"I have fun renting and selling storage."

https://stallman.org/articles/made-for-you.html

C-f Storolon

>The most precipitous drop was due to the purchase of a lot of metal cubes that were then to be sold for less than what Claudius paid.

Well, I'm laughing pretty hard at least.

The identity crisis bit was both amusing and slightly worrying.

> Can Claude run a small shop?

Good luck running anything where dependability on Claude/Anthropic is essential. Customer support is a black hole into which the needs of paying clients needs disappear. I was a Claude Pro subscriber, using primarily for assistance in coding tasks. One morning I logged in, while temporarily traveling abroad, and… I’m greeted with a message that I have been auto-banned. No explanation. The recourse is to fill out a Google form for an appeal but that goes into the same black hole into which all Anthropic customer service goes. To their credit they refunded my subscription fee, which I suppose is their way of escaping from ethical behaviour toward their customers. But I wouldn’t stake any business-critical choices on this company. It exhibits the same capricious behaviour that you would expect from the likes of Google or Meta.

Instead of dedicating resources to running AI shops, I'd like to see Anthropic implement "Download all files" in Claude.