Here’s an unpopular read on this week’s news: the consumer AI comeback is the worst thing that could have happened to consumer AI. Not because the products are bad. Some of them are genuinely fun. But popularity is the cost center. Every delighted user is a line item, and the people cheering loudest about adoption numbers are describing the size of the hole, not the size of the win.
TechCrunch made the case this week that consumer AI is back. Meta’s personal assistant Muse has been a surprise hit, along with its plush-like mascot Jolly. OpenAI dropped Dots. The vibe shift is real — after a long stretch where every interesting AI launch was aimed at a procurement department, products are being built for actual humans again.
And in the same breath, TechCrunch laid out the problem: the economics underneath these products don’t work without enterprise revenue. That’s not a footnote. That’s the whole story.
Charming products, ugly math
I review AI tools for a living, which mostly means I watch companies discover that their best product is their most expensive one. A consumer assistant that people genuinely enjoy talking to is an assistant people talk to constantly. Enthusiasm translates directly into inference bills. Unlike the SaaS era, where the marginal cost of a delighted user rounded to zero, here your most loyal users are the ones quietly draining the most money.
So the growth story has a ceiling that isn’t about demand. Consumer AI products can win attention and still hit a wall, because attention is the thing that costs money. Enterprise revenue gets floated as the fix — and it can be, in the sense that a business customer paying real money per seat can subsidize a lot of free consumer usage. But that’s not a consumer business. That’s a consumer business wearing an enterprise business as a life support system.
Ed Zitron has been making a version of this argument for a while, and his framing has aged well: the financial complexity and sheer volume of capital this industry requires to function isn’t sustainable in the medium term. You don’t have to agree with every prediction he makes to notice that the structure he’s describing matches what the product launches keep showing us.
What this means when you’re picking tools
This isn’t abstract macro talk. It changes how you should evaluate anything consumer-facing right now:
- Free tiers are marketing budgets, not features. If a product’s generosity is the main reason you like it, you’re a line item in someone’s growth experiment, and growth experiments end.
- Mascots and personality are cheap to build and expensive to run. Jolly is charming. Charm is not a moat, and charm that requires constant inference is a liability dressed as a differentiator.
- Ask who the paying customer is. If the answer is “enterprise, eventually,” the consumer version is a demo with a user base.
- Assume the pricing you see today is not the pricing you get next year. Every company in this category has an incentive to raise prices or reduce limits once the land grab cools.
None of this means you shouldn’t use these tools. I use them. It means you should hold them loosely, avoid building workflows that can’t survive a pricing change, and treat migration cost as a real factor when you adopt something new.
The pressure is coming from more than one direction
The cost side isn’t the only squeeze. Regulation is arriving in specific, operational form rather than as vague principle — a Colorado bill that took effect in February 2026 regulates high-risk AI use in sensitive sectors like hiring, housing, lending, and healthcare. That’s aimed at applied systems rather than chatbots, but it signals a direction: compliance work becomes a cost of doing business, and compliance work is exactly the kind of overhead that consumer margins can’t absorb gracefully.
Meanwhile the economists are split in ways that should make everyone uncomfortable. Reuters’ Breakingviews highlighted Oxford economist Carl-Benedikt Frey arguing that the risks to AI’s economic impact are serious, with one potential risk overshadowing the rest. Stanford’s Chad Jones has been mapping out economic futures for AI that range from abundance to apocalypse. When credentialed people describe an outcome range that wide, the honest read is that nobody knows, and the industry is spending as if the optimistic end is already banked.
Why I’m still paying attention
Consumer AI coming back is good news for anyone who thinks this technology should be useful to normal people rather than just to compliance dashboards and sales teams. I’d rather review Muse and Dots than another AI-powered meeting summarizer.
But I’m not going to pretend that adoption charts are proof of a working business. The last time a tech category got this much attention with this little unit-economics clarity, we spent a decade watching subsidized convenience slowly become expensive convenience. Enjoy the free tier. Just don’t build your life on it.
🕒 Published: