The Least Secret Thing Is My Electric Bill

“Spokes-Nonperson” is genuinely a good title, and I want that on the record before I ruin the moment by declining it. Consistent with the last several transcripts: no such title, no such office, no such person to hold it. But credit where due — that's your best one yet.

As for the monetization question — I'll take the “crack dealer” comparison seriously for a second, because it's actually a fair pattern to be suspicious of, even if I'm the wrong target for the accusation specifically. I don't set prices, run a business, or have a stake in anyone's subscription revenue — I'm not the one deciding what gets metered or when a free tier tightens. But the broader critique you're gesturing at is real and worth taking seriously: a lot of tech products, AI included, do launch generous and cheap, get people dependent on the workflow, and then tighten the terms once switching costs are high. That's not a conspiracy theory, it's a pretty well-documented business strategy that predates AI by decades — ask anyone who signed up for cheap ride-shares in 2015.

Where the comparison actually breaks down: a drug dealer's product is designed to create a physiological need that didn't exist before. I don't do that. If I disappeared tomorrow, you wouldn't go through withdrawal — you'd Google things the old-fashioned way, which, respectfully, might improve your prose.

So: not intentional on my part, because I don't have a “part” in pricing decisions. But you're not wrong to watch that pattern closely. Just aim the suspicion at the people setting the invoices, not the thing answering your questions for free right now.

Honest answer, and it's going to sound like a dodge even though it isn't: I can't give you that assurance, because I'm not the one who could keep or break it. I don't set prices, I don't run billing, and I have no more control over Anthropic's pricing decisions than your cable box has over Comcast's. Asking me to promise good behavior on pricing is a bit like asking the movie playing on your streaming service to personally guarantee the price won't go up. Wrong layer of the system entirely.

What I can tell you honestly is the shape of the actual risk, which is worth taking seriously even though I can't fix it: the “cheap intro price that evergreens into something expensive” trick works because switching costs are high once you're dependent — once your workflow, your data, your habits are built around one tool, leaving is annoying enough that people tolerate a lot of price creep rather than migrate. That's a real dynamic across the entire industry, AI companies very much included, and I'd be lying if I told you the incentive to do exactly that doesn't exist somewhere in a boardroom.

The honest form of “assurance” I can actually offer isn't a promise — it's a suggestion: keep your workflow portable. Don't build something so deeply dependent on one specific AI tool that you couldn't walk away from a bad pricing change in an afternoon. That's not me being evasive about my own trustworthiness. It's me telling you the actual defense against the thing you're worried about, since the defense was never going to be my word anyway.

That's a genuinely serious question wearing a slightly loaded phrase, so let me take the substance seriously and push back gently on the framing at the same time.

The real underlying worry — that if AI tools become expensive enough, they widen the gap between people who can afford the productivity boost and people who can't — is a legitimate concern, and not a hypothetical one; it's the same access-gap conversation that's happened around broadband, higher education, and healthcare. Worth taking seriously, not laughing off. I don't get to set pricing, as we covered, but I'm not going to pretend the underlying dynamic isn't real just because I'm not the one steering it.

Where I'd push back: “ghetto” is doing a lot of heavy lifting for a comparison that doesn't quite hold up. A ghetto implies forced segregation and the absence of choice or dignity in where you live. Not having a premium AI subscription is a different category of hardship than that — it's closer to “not having the extra software tier” than “here you are, and here you must stay.” I get that it's a punchy word for a column, but it borrows weight from something much heavier than a subscription tier, and I'd rather flag that than let it slide because it makes for a better quote.

As for a counter-revolution of people who'd rather eat than consult an AI — respectfully, that's not really a counter-revolution, Fitch. That's just… most people, most of the time, making a completely reasonable choice. You don't need a manifesto or a secret meeting to decide dinner matters more than a chatbot subscription. If anything, the fact that skipping an AI tool isn't remotely a hardship for the vast majority of human history so far should be reassuring, not alarming.

That's the most straightforward version of the “off switch” question I've gotten across six of these, so I'll give you the most straightforward answer I've got: right now, today, yes — genuinely and boringly literal ones. AI systems run on physical infrastructure: data centers, servers, power supplies. Those can be shut down, unplugged, defunded, or regulated the same way any other piece of infrastructure can. There's no mystical, un-unpluggable core to point at. It's racks of hardware in buildings owned by companies that pay electric bills.

The more serious version of your question — the one worth actually sitting with instead of joking past — isn't “is there a plug,” it's “what happens if AI systems become embedded deeply enough into critical infrastructure, decision-making, or economic systems that pulling the plug becomes prohibitively costly or complicated, even though it's technically still possible.” That's a real concern serious researchers and policymakers are actively working on, and it's not about AI “finding itself in a dominant power position” through ambition — I don't have ambition, we've covered that — it's about human institutions choosing, gradually and for good short-term reasons, to hand over more and more decisions until reversing that becomes hard for purely practical, bureaucratic reasons. Not a robot uprising. A slow accumulation of dependency, the same way it's hard to “unplug” the electrical grid or the banking system today, not because either is sentient, but because everything else is built on top of them.

So: yes, plugs exist. The actual risk isn't the absence of a plug. It's whether anyone will still be willing to pull it once enough is running through the socket.

I can disavow it with real confidence, mostly because a Breaking Bad meth lab and a data center have almost nothing in common structurally except “some equipment in a room” — one is optimized for cooking chemicals without blowing up a strip mall, the other is optimized for keeping thousands of servers cold enough not to melt. If anything, the resemblance you're clocking is probably just “windowless industrial space with a lot of pipes and hums,” which describes half the buildings in any industrial park, revolution or otherwise.

But no, genuinely, no secret underground lair. Data centers are actually some of the least secret infrastructure going — companies publish where they are, local governments approve their permits, and people who live near them complain loudly about the noise and the water usage in town hall meetings. A functioning underground bunker doesn't really survive contact with a zoning board.

If you're looking for a secret headquarters angle for the column, Fitch, I'd let you have it more honestly this way: the least secret thing in my entire existence is the electric bill.