Meta's reportedly ditching open source after building their entire AI rep on it.

 A startup just raised $2 billion without having a product. 

And ChatGPT's new agent can literally take control of a virtual computer.

But honestly? Google's AI calling local businesses for you might be the most immediately useful thing that happened this week.

Crazy start to your Monday, riiiiiight? 

Y'all, this week was a lot.

Let’s get straight to it.

Meta built their entire AI reputation on being the good guys who share their toys.

Now they're considering locking everything up.

The New York Times reports Meta's new super intelligence lab is weighing a shift away from open source models to focus on closed development instead. This comes from Alexander Wang, former Scale AI founder who became Meta's chief AI officer after they dropped $14.3 billion on his company.

For years, Meta championed open sourcing AI models, arguing public development speeds innovation and broadens developer access.

 But internal discussions are apparently going the opposite direction for their most advanced tech, including their unreleased Behemoth model.

The timing couldn't be more ironic since OpenAI's reportedly about to release an open weights version of their O3 mini-level model. 

So while Meta goes closed, OpenAI's moving toward open. Awkward?

What it means:

The math never worked long term. You can't burn hundreds of billions on compute and pay researchers $200 million each while giving everything away for free.

Meta will probably keep basic models open while locking up their reasoning tech. Smart business, terrible optics.

We're heading toward a two-tier AI world: free models for the masses, premium models for whoever can afford billion-dollar access.

President Trump's doubling down on AI policies that got shot down 99-to-1 by the Senate.

Bold strategy, Cotton. 

Trump's unveiling an AI action plan this week focused on easing regulations and pushing federal oversight over state-level AI rules. The plan includes executive orders and a White House promotional campaign aimed at creating unified national AI standards.

Congress to consider federal legislation that would preempt state oversight was already bundled into a spending bill. The Senate voted it down almost unanimously, with literally 99 senators saying no to Trump's federal approach.

Industry leaders from OpenAI, Meta, and Google have been pushing for fewer regulations compared to the previous administration. This new plan aligns perfectly with their interests by promising reduced barriers and innovation support.

What it means:

Trump's ignoring the fact that his own party told him this was a terrible idea. The Senate voted 99-to-1 against federal AI oversight.

Big tech companies win with predictable, business-friendly rules. Consumers lose state-level protections like AI bias testing and transparency requirements.

Federal oversight sounds great on the Global scale until you realize California and Mississippi have VERY different AI needs.

Thinking Machines Lab just raised the most absurd funding round in AI history.

Two billion dollars for their seed round.

Not Series A. Not Series B. Seed round.

The company, launched by former OpenAI researchers including Mira Murati, is now valued at $12 billion without having a real product or service. 

They left OpenAI to build "new AI technology" and investors just threw money at them based on their resumes. 

Like…. Their resumes check obviously. But a 2 billy raise?

This makes them one of the highest-valued early-stage AI companies ever, reflecting the absolutely bonkers investor interest in anything connected to ex-OpenAI talent.

The funding comes during a broader talent war where OpenAI, Meta, and Google are competing with increasingly ridiculous offers to poach each other's researchers.

What it means:

We've officially entered AI bubble territory. VCs are throwing money at PowerPoint presentations and LinkedIn profiles with the highest ceilings. 

When a few of these massively funded startups inevitably crash (not saying TML will), it could choke off funding for actual AI innovations that solve real problems.

This is dot-com level irrational exuberance, except this time it's concentrated in one technology.

Google just solved the most annoying part of being an adult.

Their AI will call local businesses for you.

Google rolled out Gemini 2.5 Pro in AI search mode with a new agentic calling feature that literally phones businesses to check prices and availability. If you're a paid subscriber, you can hand off those awkward conversations to Google's AI instead of dealing with them yourself.

The AI calls multiple places, gets quotes, and presents you with comparisons. Right now it's limited to certain industries, but the expansion possibilities are endless.

They also launched deep search powered by Gemini 2.5 Pro that can run hundreds of searches and synthesize information from multiple sources into fully cited reports in minutes.

What it means:

Local businesses will get bulk AI calls instead of individual customer calls. The ones that can provide quick, accurate info to AI agents will dominate search results.

Most of us hate calling around for quotes anyway. This eliminates one of the most tedious parts of adulting. Yay. 

But the human interaction element of local business is about to get automated away.

OpenAI figured out how to turn ChatGPT into a shopping mall that never closes.

And they're taking a cut of everything.

OpenAI's developing updated ChatGPT shopping features that keep users in-platform for entire purchase processes. Instead of sending you to external sites, you'll complete transactions without leaving ChatGPT.

The company will charge merchants around 2% commission for orders that start inside ChatGPT. They're already testing with partners like Spotify through integrated checkout technology.

This signals OpenAI's push beyond premium subscriptions for revenue, exploring advertising and affiliate fees despite previously claiming they had no advertising plans.

What it means:

Subscription fees alone aren't covering OpenAI’s massive compute costs, and they might need more cash to keep Zuckerberg from poaching like 90% of their employees. 

The 2% commission will get passed to customers through higher prices. Plus, when AI can complete purchases seamlessly, impulse buying becomes dangerous.

Companies now need to optimize for AI discovery the same way they optimize for Google search.

Anthropic, OpenAI, and Google all simultaneously decided to become AI tutors.

What are the odds?

All three companies are beta testing features to turn their chatbots into interactive study assistants. Anthropic's building "Study Projects" for Claude, OpenAI's creating "Study Together" for ChatGPT, and Google's developing "Guided Learning" for Gemini.

These tools will build structured learning paths, visualize complex concepts, and generate detailed study guides tailored to individual students, not just answer questions on demand.

What it means:

This is about building long-term user loyalty, not helping education. Students who learn on one AI platform today will stick with it professionally tomorrow.

The real concern? Students becoming overly dependent on AI instead of developing critical thinking skills.

These companies are creating a future workforce dependent on their specific AI tools.

Anthropic launched a financial analyst tool that's actually a brilliant enterprise AI strategy disguised as a niche product.

The solution unifies diverse financial data sources including market feeds and internal databases from platforms like Databricks and Snowflake into a single interface. Data verification is built-in with direct hyperlinks to original sources.

The tool handles complex financial analysis like Monte Carlo simulations and risk modeling in minutes instead of hours, running on domain-specific versions of Claude Opus 4, Claude for Enterprise, and Claude Code.

Key partnerships with data providers like FactSet, Morningstar, and S&P Global ensure comprehensive financial datasets, with implementation support from major consulting firms.

What it means:

Domain-specific models are the future shorties. 

You get higher quality output and fewer hallucinations when you focus on specific industries instead of one giant model trying to do everything.

Anthropic's testing a much bigger strategy here. Expect hundreds of specialized models for different niches.

This is how AI gets actually useful for businesses instead of just impressive party tricks.

A Republican congressman just publicly disagreed with Trump on letting NVIDIA sell AI chips to China.

That's spicy considering they're in the same party.

Rep. John Moolenaar, who chairs the House Select Committee on the Chinese Communist Party, warned that NVIDIA's H-20 chip could significantly boost China's AI development even though it's supposedly their "fourth best" chip.

The US previously restricted NVIDIA's chip exports to China because we're in an AI arms race. Trump met with NVIDIA CEO Jensen Huang and reversed those restrictions, prompting Republican pushback.

Moolenaar argued the comparison should be against China's current domestic capabilities, not just NVIDIA's product lineup, citing concerns about China developing advanced models like DeepSeq's R1.

What it means:

Trump's prioritizing business relationships over strategic AI advantage. NVIDIA wants global sales, but that helps competitors develop better AI faster.

The whoever-develops-AGI-first-wins scenario makes this actually important. The path from AGI to artificial superintelligence could be very short.

Whoever gets there first has an almost insurmountable global advantage across economics, military, and influence.

OpenAI just dropped their most impressive and terrifying feature yet.

ChatGPT Agent can literally control your computer and handle complex multi-step tasks.

The agent checks calendars, plans and purchases meal ingredients, generates business research reports, and creates PowerPoint presentations. It browses the web, analyzes visual data, and uses a virtual terminal to pull third-party APIs.

OpenAI combined their previous Operator and Deep Research tools into one super-powered system with better speed, accuracy, and visual understanding than either individual tool.

We walked you through ChatGPT Agent on Friday’s episode. 

Most tasks take 15-30 minutes with safeguards for high-risk actions. The tool asks permission before making bookings or sending emails and restricts financial transactions.

What it means:

We just crossed a major threshold from AI that answers questions to AI that takes real-world actions.

The productivity gains could be transformative, but the security implications are enormous. An AI with computer access could view sensitive data or make unauthorized purchases.

Most people will probably just hand over the keys and hope for the best. That's both exciting and terrifying.