🤖 Zuckerberg admits $100B mistake

Zuckerberg admits the mistakes. The market's still waiting for the wins.

June 15, 2026

My fellow AI explorers

It's been one of those weeks where the AI industry feels less like a tech sector and more like a soap opera with quarterly earnings calls. Zuckerberg is publicly admitting Meta screwed up its AI reorg. A US government order meant to slow down China's AI ambitions just sent two Chinese AI stocks into orbit. And the UK government decided the thing kids really need protecting from isn't just social media, it's AI boyfriends and girlfriends.

Buckle up, this one's a ride.

In today’s edition:

Meta Business

A Year After Hiring Alexandr Wang for $14 Billion, Meta Is Still Figuring Out What to Do With Him

Mark Zuckerberg told employees in an internal memo that Meta has "made mistakes" in its AI workforce transformation, and honestly, the timing could not be more on the nose. This comes almost exactly a year after Meta paid roughly $14.3 billion to bring in Scale AI's Alexandr Wang and a wave of his top engineers to lead a "strategic rebuild" of Meta's AI stack.

So what did that $14 billion buy, you ask?

  • Wang's team shipped Muse Spark in April, Meta's first proprietary foundation model and a real break from its open-source roots

  • Meta's stock is now underperforming compared to every other tech megacap, despite the spending spree topping $100 billion in AI infrastructure

  • Internally, Meta carried out a brutal restructuring in May, laying off 10% of its global workforce and reassigning 7,000 employees into AI-focused roles

  • Some of Meta's new AI units reportedly ran with a 50:1 ratio of individual contributors to managers, a structure so flat it apparently caused its own headaches

In the memo, Zuckerberg admitted the company "will almost certainly make more" mistakes, but tried to reassure staff that no further company-wide layoffs are planned this year. He's also promising more team-building budget and a company-wide hackathon in July, which, sure, nothing says "we fixed the culture problem" like a hackathon. Everybody knows that…

Here's the part that should make founders and operators pay attention.

Wang was hired to fix the technology. He delivered a model. But analysts are now openly saying Meta's real problem isn't research talent; it's productizing and selling what gets built. That's Zuckerberg's job now, and the market is watching closely to see if he can do it.

Meta still pulls 98% of its revenue from ads. If Muse Spark and whatever comes next can't move that number, this entire $100 billion bet starts to look like the most expensive science project in tech history.

🔮 Prediction: Expect Meta to lean hard into consumer-facing AI features across Instagram, WhatsApp, and the standalone Meta AI app over the next two quarters, because that's the fastest path to "proof of adoption" that investors are demanding. Watch for a major Meta AI feature announcement timed deliberately around an earnings call.

What do you think? Is this a temporary growing pain for Meta, or a sign that throwing money at talent doesn't fix a product culture problem? Hit reply, I read every single one.

AI Stocks

A US Export Ban on Anthropic Just Sent Chinese AI Stocks Soaring 33%

This is the kind of story that sounds like it was written by an AI with a sense of irony, except it's completely real.

The Trump administration ordered Anthropic to suspend access to its most advanced models, Fable 5 and Mythos 5, for any foreign national, including Anthropic's own non-citizen employees. He cited national security concerns. The order reportedly followed warnings that Fable 5 could be jailbroken, and concerns about a China-linked group accessing the model.

The market's reaction was immediate, and not in the direction Washington probably hoped for.

The strategic logic here is almost too clean. Washington tries to lock down frontier AI access on national security grounds. The immediate effect is that any company outside the US that was relying on, or considering relying on, Anthropic's top models now has a very real reason to hedge with a second source. Chinese "AI tigers" like Zhipu and MiniMax are positioned to be exactly that second source, and they're framing themselves as the open, reliable, no-strings-attached alternative.

Whether or not that's fair to Anthropic, who didn't choose this policy, the optics are brutal. A move designed to slow down the global AI competition just became a marketing campaign for the companies it was supposedly meant to contain.

🔮 Prediction: This won't be the last export restriction that backfires this way. Expect more enterprises outside the US, especially in regions wary of getting caught in US-China tech tensions, to start quietly building multi-model strategies that include at least one Chinese open-source option, regardless of how this specific policy plays out.

Do you think export controls like this actually slow China down, or do they just accelerate the exact outcome they're trying to prevent? Reply and tell me, I want to hear the counterargument too.

30-Second AI Play

Want to go from "what even is an AI agent" to having a working one on your machine? Google's Agent Development Kit (ADK) makes this surprisingly approachable, even if you've never touched agent frameworks before.

  1. Understand the core loop. Every modern AI agent follows a simple cycle: reason about the task, take an action like calling a tool or API, observe the result, then decide what to do next. This is the same idea behind the ReAct framework, and it's the foundation of basically every agent you'll build.

  2. Set up your environment. Install UV (Python's fast package manager), then install Google's ADK. Create a new project file and import the ADK libraries you'll need.

  3. Build a planner agent. Give it one job: take a topic and turn it into a structured outline with a title, intro, four to six sections, and a conclusion. Set its output to save into the shared state so other agents can use it.

  4. Add a validation loop. Wrap your planner in a "loop agent" alongside a checker agent that reviews the output and either says OK or retry. This gives your agent up to three attempts to get it right automatically.

  5. Connect a writer agent and a root agent. The writer takes the validated outline and turns it into a full draft. The root agent (call it your "manager") only has access to the planner and writer as tools, so it stays clean and controlled. Run it with adk web to interact through a local browser UI.

💡 Pro Tip: Start with a narrow, low-stakes use case like a blog post generator before building anything customer-facing. The validation loop pattern (write, check, retry up to three times) is the single most useful concept here, because it's what turns a flaky AI output into something you can actually trust in production.

Want to get your business in front of 150k+ startup founders, engineers, and AI early adopters?

Other Relevant AI News!

🤝 OpenAI launched the OpenAI Partner Network, a $150 million program to help consultancies like Accenture, McKinsey, BCG, Bain, and PwC deploy AI inside enterprises, with a goal of training 300,000 certified consultants by the end of the year.

💰 Mistral AI is reportedly in early talks to raise around $3.5 billion at a $23 billion valuation, more than double its valuation from last September, as the French startup pushes deeper into enterprise and government AI infrastructure across Europe.

👶 The UK government confirmed it will ban social media for under-16s by 2027 and separately require AI "romantic companion" chatbots designed to simulate relationships to enforce a minimum age of 18.

Golden Nuggets

  • 🧑‍💼 Meta's $14B AI hire delivered a model, but the real challenge is now Zuckerberg's job to sell it

  • 🇨🇳 A US export ban on Anthropic accidentally became free advertising for Chinese AI labs, sending Zhipu up 33%

  • 🌍 OpenAI is going all in on enterprise consulting, while Mistral chases a $23B valuation as Europe doubles down on AI sovereignty

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI