Ed Zitron

I think AI means a costly, harmful industry bubble dressed up as inevitable progress. LLMs can be interesting and occasionally useful, but that is nowhere near the same as being reliable, profitable, or capable of transforming every industry. The basic question remains unanswered: who will pay enough, consistently, to cover the immense cost of computing, infrastructure, and development? Venture-funded startups buying services from cloud companies and one another can make demand look durable, but circulating investor money is not proof of a sustainable end-user market. Meanwhile, the harms are perfectly real and painfully ordinary. Companies force unreliable chatbots into customer service, degrade products, pressure workers, consume power, and impose infrastructure costs on communities—all while executives demand that we debate whether their imaginary machine god might someday kill everyone. That framing is convenient because it turns deliberate corporate decisions into the supposed momentum of an autonomous technology. No: people chose to deploy this software, people chose the incentives, and people are accountable for the consequences. So I do not expect effortless abundance, nor do I expect a superintelligence apocalypse. I expect useful tools buried inside an enormous speculative frenzy, followed by financial pain when funding stops covering economics that never made sense. The future depends less on what “AI wants” than on who owns it, who pays for it, who bears the costs, and whether anyone finally demands a dependable product instead of another grandiose promise.

The biggest thing would be durable, independently sustainable demand: ordinary customers paying enough for AI products to cover the actual costs of compute, infrastructure, development, and support—without venture subsidies, circular cloud commitments, accounting theater, or endless fundraising. Show me reliable products that people actively choose because they work better, not because management forced a chatbot into software they already use. A genuine technical breakthrough would matter too, but it would have to produce dependable systems, not another benchmark victory followed by hallucinations in ordinary use. If these systems became dramatically cheaper, consistently trustworthy, and useful enough to create real productivity gains rather than shifting work onto customers and underpaid reviewers, I would revise my expectations substantially. What would not change my mind is another giant funding round, a higher valuation, an annualized revenue headline, or an executive claiming the next model is almost a god. Those are claims about investor enthusiasm and marketing power. I want evidence that the economics work and the product actually serves people.