01The fake Mark
Meta is building an AI Mark Zuckerberg.
Voice, face, how he talks, how he thinks. Made by the Superintelligence Labs team. Meta says it’s for all 79,000 workers — so anyone can talk to fake Mark and get a fake-Mark answer. Meta frames it as “staying in touch.”
Jim’s take
A fake Mark can say “you’re fired” to a thousand people at the same time. Far more efficient. And we know how much he loves efficiency. The good news for Meta: they already have a lot of practice on what firing people looks like.
02The product update
Claude Code, now with routines.
Anthropic shipped a bigger, better version of the Claude Code app. Run more than one job at a time. Drag things around. And routines: set up a task once and have Claude run it every morning on its own. You don’t have to be there.
This matters because Anthropic is no longer just selling chatbots. They’re selling a service — one that runs on your machine, on your schedule, while you go about your day.
03The pricing pivot
Billed by the minute.
Anthropic changed how they charge enterprise customers. It could triple costs. And I’m pretty sure I know why.
The old way — simple
One flat fee. As much Claude as you could use.
- ModelFlat monthly fee per worker.
- UsageEffectively unlimited.
- PredictabilityHigh. Finance teams liked it.
- FeelingSalaried.
The new way — metered
Per-seat and per-use. Every action bills.
- ModelFlat seat fee + usage fees on top.
- UsagePriced by how much work the AI does.
- PredictabilityVariable. Bills climb with adoption.
- FeelingBillable hours.
Every bit gets built on top. It just keeps adding up.
04Why now
The IPO math.
Internal papers say Anthropic is on track to lose $14 billion this year. Reports have them targeting an IPO as early as October — about six months out. You cannot go public if your company is losing billions with no plan to stop the bleeding. This isn’t the dot-com boom.
So Anthropic is changing prices now, well before they have to show their books to the world. Angry customers will complain today. By October, it’s old news.
Jim’s take
Smart timing, Anthropic. But as costs climb, look for enterprises to shift some of the load to the cheaper models based in China. The pricing pivot only works if the alternative is worse. It’s getting less worse.
06The hacker study
The criminals don’t trust it either.
A new study read more than 160 chats across 170 hacker sites — nearly 700,000 posts over seven months — and found something the headlines missed: hackers aren’t really using AI all that much. Mostly, they’re curious.
In roughly 18% of chats, hackers complained the AI was making mistakes and that the code it writes is buggy. They said it floods their forums with junk. And those scary “WormGPT” and “FraudGPT” products in the news? The hackers themselves say they’re fake — regular ChatGPT with a new label slapped on.
The same reason your company is slow to use AI is the same reason criminals are slow to use it. Nobody trusts it yet. Not you, not your boss, not even the bad guys.
