Nuclear Waste and AI Agents: Two Problems We Can’t Afford to Ignore

Nuclear energy is enjoying a rare moment in the sun. Public approval is up, Big Tech is throwing cash at it to feed their insatiable electricity demand, and politicians on both sides of the aisle are suddenly fans. Great, right?

Except there’s this old problem nobody wants to talk about: the waste.

Every year, US nuclear reactors churn out about 2,000 metric tons of high-level radioactive waste. And we have no permanent home for any of it. That’s not a new problem—it’s been festering for decades. But with new reactors being proposed and the old ones still running, the urgency is real. Casey Crownhart over at MIT Tech Review lays out the stakes in a recent piece, and it’s worth reading if you care about where all that spent fuel ends up.

I’ve been following this space for years, and the frustrating part is that the technical solutions exist. Deep geological repositories, advanced reprocessing, even some next-gen reactor designs that burn waste as fuel. The holdup is political, not technical. Yucca Mountain was supposed to be the answer, and we all know how that turned out. Finland is actually building a permanent repository, but the US is still dithering. If we’re serious about nuclear as a climate solution, we need to get serious about the back end.

On the AI front, Will Douglas Heaven has a piece that cuts through the hype to talk about what actually matters: AI agents that work together. Not the chatbots that can hold a conversation, but systems that can actually do things—book meetings, manage supply chains, handle customer service flows.

The real leap, he argues, comes when you have multiple agents coordinating. Think of it like an assembly line for knowledge work. One agent gathers data, another analyzes it, a third writes the report, a fourth double-checks for errors. That’s the vision, and apps like Codex and Claude Cowork are starting to show what it looks like in practice.

But here’s the thing I don’t see enough people talking about: the failure modes multiply when you have multiple agents in play. One rogue agent hallucinating a fact is bad enough. A whole team of them passing around bad data and reinforcing each other’s errors? That’s a recipe for catastrophic decisions. Heaven touches on the risks, and I think we’re going to see some spectacular failures before we get this right.

Agent orchestration made MIT Tech Review’s list of 10 Things That Matter in AI Right Now, and that’s a fair call. But I’d add a caveat: it matters because it’s going to break things before it fixes them.

Also worth your time: Stephen Ornes has a fascinating piece on “mirror bacteria”—lab-created microbes with reversed molecular chirality. The idea was to unlock new insights into cellular biology and drug design. But now many of the same scientists who proposed the research are warning that these organisms could pose an existential threat. If a mirror bacterium escaped, our immune systems might not recognize it, and antibiotics might not work on it. It’s the kind of scenario that keeps biosecurity experts up at night. The full story is available as an MIT Tech Review Narrated podcast, which is worth a listen if you prefer audio.

And in the must-reads: Elon Musk testified for the first time in the OpenAI trial, claiming Sam Altman “stole a charity.” The legal showdown is getting juicy, and it’s a reminder that the AI world’s drama is only heating up.