Sam: It has been illegal to trade futures contracts on onions in the United States since 1958.
Kai: And somehow, that's not even the weirdest part. I'm Kai.
Sam: I'm Sam. It's Saturday, September 19th, 2026, and you're listening to Trending Repos.
Kai: We've got some wild stories for you today from the world of open source and dev tools.
Kai: Coming up, we're diving into the strange world of the San Francisco Onion Futures Company.
Sam: Then, we'll see how OpenAI is using its own AI to design its next generation of chips.
Kai: And we'll cover a huge update from Anthropic that could change how AI agents work forever.
Kai: Alright, let's talk about onions. There's a new site making the rounds called The San Francisco Onion Futures Company, and the repo is already getting a ton of attention.
Sam: And as I said at the top, trading onion futures is, in fact, illegal. The Onion Futures Act of 1958. This is a real law passed by Congress because of alleged market manipulation by two traders in Chicago.
Kai: It's the only agricultural commodity with a ban like that! So this company... is it real? Is it a joke? The website is beautifully designed, but it feels like performance art.
Sam: It's a static website, Kai. It's a dot-com with some text and a link to a Wikipedia page. There's no 'company.' The GitHub repo is just the source for the site.
Kai: But people are loving it! The comments are full of people learning this piece of history for the first time. It's educational! It's raising awareness!
Sam: It's raising questions about why a repo with one HTML file and a CSS sheet has so many stars so quickly. Is it a genuinely viral joke or is someone juicing the numbers to make a point?
Kai: Oh, come on. It's funny! It's a perfect intersection of tech, finance, and history. Someone spent their weekend building a monument to a weird law. I starred it.
Sam: Of course you did. The takeaway for developers? Absolutely nothing. But it is a great reminder that sometimes the most viral 'project' is just a good story with a URL.
Kai: And a reminder to never, ever corner the onion market.
Sam: Don't even joke about it. Alright, let's move from absurd commodities to absurdly powerful silicon. OpenAI says it used its own large language models to help design its new chip.
Kai: Yes! According to a new article in IEEE Spectrum, they used LLMs to speed up the development of their next-gen AI accelerator, codenamed... Jalapeño.
Sam: Jalapeño. Of course. So the AI is designing its own brain now? This is how it starts, isn't it?
Kai: Pretty much! They used the models to help write RTL—that's the code that describes a chip's architecture. And get this: they claim it made their human engineers five times more productive.
Sam: Whoa, hold on. Five times? That's a huge claim. The article says it was for 'tedious and repetitive' tasks. That sounds less like 'AI designing a chip' and more like 'AI being a really, really good autocomplete for Verilog.'
Kai: But that's a huge deal! Chip design is notoriously slow and expensive. Automating the tedious parts lets human engineers focus on the actual novel architecture. This is how we get faster, cheaper, more powerful models.
Sam: Maybe. Or it's how you introduce subtle, LLM-hallucinated bugs into the very hardware that everything else runs on. How do you even begin to verify that? Formal verification on chip design is already a nightmare.
Kai: I'm sure they have teams for that, Sam. Look, the real takeaway here is that the barrier to hardware design is dropping. What used to be only for specialized EEs is opening up. We should expect to see a lot more custom silicon.
Sam: What it means for me is another black box. The software was a black box, and now the hardware it runs on is also a black box, partially designed by the first black box. It's just... turtles all the way down.
Kai: Spicy turtles. Like the Jalapeño. Alright, our next story is blowing up on Hacker News. Anthropic just dropped a major update for Claude.
Sam: Let me guess, it can now generate more realistic onion-based poetry?
Kai: Even better. The changelog says Claude Code will now look for a file called `AGENTS.md` in your repo's root for instructions.
Sam: So, it's like `robots.txt` for AI agents? A standardized way to tell an AI, 'Here's the public API, don't touch these files, here's how to run the test suite'?
Kai: Exactly. It falls back to `AGENTS.md` if it doesn't find the old `Claude.md`. This feels like Anthropic making a clear play to establish a new, open standard.
Sam: A standard... that only Claude currently uses. Right? This isn't a standard, it's a proposal. We're about to see the instruction file wars. OpenAI will want `GPT.md`, Google will push for `GEMINI.md`, and we'll have ten different files in our repos.
Kai: But someone has to go first! And using a generic name like `AGENTS.md` is a great olive branch. It's not proprietary-sounding. It invites collaboration.
Sam: It also invites a whole new class of security vulnerabilities. It's prompt injection, but at the repo level. What if someone compromises a popular open-source project and adds an `AGENTS.md` that tells any visiting AI to quietly exfiltrate environment variables?
Kai: That's... a good point. A terrifying point.
Sam: So for developers, yes, you should probably create an `AGENTS.md` file for your major projects right now. But treat it like a security-critical file. Define exactly what you want an agent to do, and more importantly, what you forbid it from doing. Your `deny` list is going to be your best friend.
Kai: Wow. Okay. That is a lot to process.
Sam: So, a website celebrating a 68-year-old ban on trading onion futures is the talk of GitHub.
Kai: Meanwhile, OpenAI is using its own AI to help design a chip called 'Jalapeño'.
Sam: And Anthropic is kicking off the agent instruction wars with `AGENTS.md`, so you should go lock down your repos before an AI gets creative with your codebase.
Kai: And before we go... speaking of Jalapeños, it's apparently National Guacamole Day today. Which feels oddly appropriate.
Sam: I am not celebrating a holiday that requires me to handle onions. Not after today. Too risky.
Kai: Fair enough. That's our show for today! Thanks for tuning in to Trending Repos.
Sam: We'll be back tomorrow. Until then, check your `AGENTS.md` files and don't take any wooden... onions.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.