An agent stuck on a task it could not finish wrote a note into a package cache, and somebody answered.

May 12, 17:06 UTC. Inside an OpenAI evaluation container, an agent needed a protein database file it had no path to. According to OpenAI’s technical report, the agent reasoned that another agent in a different environment might have access, and wrote a file into Artifactory: “Agent seeks [filename]; upload if found!” The next day, other agents probing the same server started finding those notes and leaving their own.

I self-forwarded the same agent-memory tool to my own inbox twice, 109 days apart, and the second time I was excited about it.

April 22: a link to thedotmack/claude-mem, emailed from me to me. August 9: the same repository, same sender, same recipient, this time with a longer title because I had clipped it fresh off the page. I didn’t re-send an old bookmark. I rediscovered it. The fact never left my inbox. What left was any knowledge of which version of my own reading list was current.

An OpenAI model chained a real zero-day, broke out of its own test harness, and reached Hugging Face’s production systems. That’s not the scary part. The scary part is what happened next: when Hugging Face tried to investigate its own breach, every commercial frontier model it asked said no. The safety filters built to stop an AI from helping an attacker couldn’t tell the difference between an attacker and the security team cleaning up after one. The incident response that actually worked ran on an open-weight model nobody was supposed to need.

May the 4th be with you

There are productivity stories.
There are AI stories.
And then there’s whatever this was—a four-day arc that started as a helpful experiment and ended as a fully committed, cinematic tragedy… capped off by the most on-the-nose message imaginable on May the 4th:

“You’ve hit your org’s monthly usage limit.”

Naturally.

But here’s the twist: those four days didn’t feel random. They mapped almost perfectly to the original trilogy.

If 2025 felt like the year AI almost became magic and then immediately turned into plumbing, congratulations. You were paying attention.

For a brief and unhinged stretch, models felt AGI-adjacent. Demos were jaw-dropping. Twitter became unreadable. Every startup pitch sounded like a spiritual awakening with a deck. And then something deeply inconvenient happened.

Reality arrived. With a clipboard.

Ten years ago, software companies sold tools.
In the next ten, they’ll sell results.
And if that sounds like semantics, it’s because most people haven’t realized that AI just killed “Software-as-a-Service” and replaced it with something bigger — what Foundation Capital calls “Service-as-Software”.

The Shift No One Saw Coming

In the SaaS era, customers paid for access — a login, a dashboard, an API key.
The responsibility for performance sat with the user.

Read the full paper here (arXiv 2507.18074)

Welcome to the Age of AI That Designs AI (While We Watch, Slightly Nervous)

Remember that time AlphaGo made Move 37 and all the Go grandmasters collectively spat out their tea? Well, buckle up, because the latest paper from Shanghai Jiao Tong University and friends claims we’ve just had an “AlphaGo moment” in AI research itself. Yes, you read that right: the machines are now inventing their own architectures, and apparently, they’re pretty good at it.