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

Two stories caught my eye today, and they’re less disconnected than they seem. One is about a problem we’ve been kicking down the road for decades. The other is about a shiny new thing that everyone’s excited about, but nobody’s really thought through yet.

Let’s start with the old problem.

Nuclear waste isn’t going anywhere

Nuclear energy is having a moment. Public approval is up, politicians on both sides like it, and Big Tech is writing checks to power data centers with reactors. That’s all fine. But here’s the thing nobody wants to talk about: the waste.

Every year, U.S. reactors produce about 2,000 metric tons of high-level waste. And we still don’t have a permanent place to put any of it. Not a single site. Not even a plan that’s gone anywhere.

This isn’t new. Yucca Mountain was supposed to be the answer, and that’s been dead for years. Now with new reactors being discussed, the pile is only going to grow. Casey Crownhart over at MIT Tech Review makes the point well: the renewed interest in nuclear is exactly why we need to finally deal with this.

I’ve been watching this cycle repeat for years. Nuclear gets popular, everyone promises to solve the waste problem, then interest fades and the waste stays in dry casks and cooling pools. This time feels different only because the demand is coming from tech companies who are used to getting what they want. Maybe they’ll actually push for a solution. Maybe.

AI agents are coming for your job—in teams

The other big story is about AI agents. Not the chatbots that answer questions, but the ones that actually do stuff. Will Douglas Heaven has a piece on orchestrated agents, and it’s worth paying attention to.

ChatGPT proved AI can talk. That’s impressive, but it’s not transformative. The real shift comes when agents work together in teams, coordinating multiple roles to handle complex tasks. Think of it like an assembly line for white-collar work.

Apps like Codex and Claude Cowork are early examples. They’re not perfect, but they show where this is heading. A network of agents could handle research, drafting, analysis, and decision-making across an entire workflow.

The vision is compelling. But the risks are real too. When agents start interacting with real systems—sending emails, updating databases, making purchases—the margin for error shrinks fast. One misconfigured agent could cause a mess that takes weeks to clean up.

I’ve seen this pattern before in software automation. The first wave is always too optimistic, then reality hits, then we figure out the guardrails. We’re still in the optimistic phase with multi-agent systems.

The mirror life warning

There’s a third story here that’s worth mentioning, because it’s the kind of thing that keeps me up at night. Scientists who once pushed for funding to create “mirror” bacteria—life forms built with reversed molecular chirality—are now warning that it could be catastrophic.

These organisms would be invisible to our immune systems, resistant to antibiotics, and potentially unstoppable if they escaped the lab. The same researchers who wanted to build them are now saying: don’t.

It’s a rare example of scientists publicly reversing course on their own work. That takes guts. And it’s a reminder that the most dangerous technologies aren’t always the obvious ones.

What ties this together

All three stories are about the gap between what technology can do and what we’re prepared to handle. Nuclear waste is a solved technical problem—we know how to store it—but the political and social will has never been there. AI agents are technically promising, but we’re still figuring out the safety and coordination issues. Mirror life was technically feasible, until people realized the consequences.

That’s the pattern. We’re great at building things. We’re terrible at planning for what happens after.

Elon Musk testified yesterday in the OpenAI trial, claiming Sam Altman “stole a charity.” That’s a whole other circus, but it fits the theme. Even at the highest levels of tech, nobody’s really thought through the long-term implications. They’re too busy building the next thing.

I don’t have a neat conclusion here. Just an observation: the problems we ignore don’t go away. They just wait until we can’t ignore them anymore.