OpenAI grabbed so many headlines this week you’d swear they were a corduroy pillow factory.

Oh, and the ChatGPT parent company company just became the most valuable private company in the world at a $500 billion valuation while striking massive chip deals with AMD.

Meanwhile? Anthropic dropped Claude Sonnet 4.5 claiming it's the best coding model in the world and Meta wants to use your AI chat data for SUPER personalized ad targeting in their social platforms.

That couldn’t get awkward, right? lolz.

And a new Senate report warns AI could eliminate 100 million US jobs in the next decade.

What an absolutely massive week of AI news.

Here's what matters y’all. No B.S., just what you need.

OpenAI just handed 800 million people the ability to build AI agents without writing a single line of code.

(Hours after our livestream, which teased this before the announcements.)

AgentKit launched at Dev Day today, featuring Agent Builder—a visual canvas for composing agent workflows with drag-and-drop nodes, according to Sam Altman's keynote.

No code required.

Users can start from predefined templates like customer service bots, data enrichment pipelines, and document comparison flows to rapidly assemble solutions.

The Canvas offers modular building blocks. Logic nodes for if-then connections. Connectors like MCP protocol. User approvals. Built-in visual guardrails.

AgentKit also includes ChatKit, which provides a simple embeddable chat interface that developers can customize with their own branding and workflows.

Here's where it gets wild though y'all.

OpenAI engineer Christina Huang built an entire AI workflow and two AI agents live onstage in under eight minutes to prove how easy it is.

What it means: This is OpenAI's play to make agent building accessible to non-technical users, not just developers who can code.

If you've been wanting to automate workflows but didn't want to learn API integration, this could actually let you do it.

The real test will be whether these drag-and-drop agents can handle production workloads or if they're just fancy prototypes.

That screenless AI device from OpenAI and Jony Ive? Major problems.

The Financial Times reports the teams are encountering significant technical hurdles that could delay their planned 2026 launch of the palm-sized AI device.

The device was designed to take continuous audio and visual cues from the environment and respond without a traditional screen.

Always on. Always listening.

And that's exactly the problem. The team struggles to ensure the assistant engages only when truly helpful and ends conversations at appropriate times, according to sources familiar with the project.

Critical for privacy and usability.

The project also faces unresolved questions about the device's personality—how it expresses tone, initiative, and boundaries.

Privacy handling has been a central concern. How will continuous sensing protect user data? What gets processed locally versus in the cloud? Who has access?

But here's the bigger issue y'all.

One person close to Ive told the FT that compute is another huge factor for the delay. Amazon has the compute for Alexa. Google has it for Home. But OpenAI is struggling to get enough compute for ChatGPT, let alone millions of AI devices.

What it means: The always-on listening feature is exactly what's causing the delays—they can't figure out when the device should shut up.

Privacy concerns around continuous recording are legitimate and could kill this product before it launches if they don't nail it.

The compute issue is the bigger red flag—if OpenAI can't handle ChatGPT's compute needs reliably, adding millions of devices to that load is a massive infrastructure problem.

You didn’t sleep on this one, did you?

Yep, Perplexity just released Perplexity Comet to everyone for free. No wait list. No invite codes. Just download and go, according to Perplexity's announcement.

Got buried in the news cycle, but this is actually pretty significant.

The browser is called Comet and it's currently available for desktop only. Windows 10 and 11. MacOS with Apple M1 or newer chip.

Comet's key feature? A built-in AI assistant that lets users ask questions, assign tasks, and interact directly in the browser without opening a separate chatbot tab or app.

Here's where it gets interesting though y'all.

When you type a query in the address bar, it opens Perplexity instead of Google search. They're really looking to cut into Google's browser share since Chrome currently commands nearly 72% of global browser market share.

Important thing to know. Comet is built on top of Chromium. So you can actually sign in with your Chrome credentials and everything will be synced—bookmarks, extensions, everything.

Same with Microsoft's Edge browser.

What it means: Perplexity's positioning itself to capture search traffic at the browser level, not just the search engine level.

If you're a Chrome user, switching to Comet will feel seamless since it's built on Chromium.

The real question is whether native AI assistance is compelling enough to break Chrome's dominance, and right now the answer is probably no.

How many new millionaires just got minted in San Francisco?

Bloomberg reports that OpenAI facilitated a $6.6 billion share sale last week held by current and former employees, pushing the company's valuation to $500 billion—the highest ever for a privately held company.

Purchasers included SoftBank, Dragoneer Investment Group, and Thrive Capital.

This was a secondary transaction, meaning proceeds went to insiders rather than OpenAI itself. But it serves as a major retention tool amid tightening competition for AI talent.

OpenAI's last primary funding round closed in August at a $300 billion valuation, so this secondary sale pushed the new valuation up $200 billion in just two months.

The previous most valuable private company was SpaceX at a $400 billion valuation.

Even compared to public companies, this puts OpenAI in the top 20 companies in the US, ahead of Netflix that's been around for decades.

What it means: Current and former OpenAI employees who held their shares just became extremely wealthy overnight.

This is a retention move—when you're suddenly worth millions, it's harder to leave for a competitor.

The path to going public is clearer now with the recent Microsoft agreement setting groundwork for conversion to a for-profit entity.

Starting December 16th, every conversation you have with Meta AI becomes ad targeting data.

According to Meta, conversations with Meta AI about topics like travel or shopping will inform what reels, posts, and ads show up in your feed across Facebook, Instagram, and Messenger.

User notifications roll out tomorrow.

Meta AI now has more than 1 billion monthly active users, making this change pretty significant.

Voice interactions through Ray-Ban Meta smart glasses will also feed into Meta's recommendation engine.

Users cannot opt out of this recommendation change. If you don't use Meta AI, the updates won't apply to you—that's your only option.

WhatsApp interactions won't be included unless you link your WhatsApp account to sibling apps like Instagram.

What it means: If you use Meta AI, your conversations are now ad targeting data with no opt-out and no choice.

The change becomes automatic starting mid-December.

The only way to avoid this is to stop using Meta AI entirely.

Anthropic just declared Claude Sonnet 4.5 the best coding model in the world.

Released last week, the company claims state-of-the-art performance on coding benchmarks and the ability to run autonomously for over 30 hours on complex projects.

One of the biggest features is improved agentic capabilities. Anthropic says Sonnet 4.5 maintains focus for extended periods on complex multi-step tasks—reportedly over 30 hours in some enterprise trials.

The coding performance is where Anthropic has carved its niche in software development.

Anthropic claims industry-leading results against benchmarks like SWE-bench Verified. The model generates higher-quality code, is better at identifying code improvements, and can follow instructions more reliably than previous versions.

Pricing remains the same as Claude Sonnet 4 at $3 per million input tokens and $15 per million output tokens.

What it means: If you're using AI for coding, this is currently the top option based on benchmarks.

The 30+ hour autonomous coding capability is significant for building complete applications without constant human oversight.

But OpenAI had Dev Day today and Google could drop Gemini updates any time, so this "best in the world" title probably won't last long given how fast the space moves.

AMD's stock just had its biggest rally in nearly a decade.

The chipmaker's shares soared as much as 25% in premarket trading after OpenAI and AMD struck a multiyear partnership to roll out AMD Instinct AI chips at massive scale, according to a joint statement released this morning.

OpenAI will deploy six gigawatts of AMD GPUs over multiple years, with the first gigawatt deployment starting in the second half of 2026 using AMD Instinct MI450 GPUs.

AMD also granted OpenAI the right to buy up to 160 million AMD shares if OpenAI hits deployment and stock price targets, which could give OpenAI about a 10% stake in AMD.

Some estimates place the total revenue for AMD in the tens of billions of dollars over the next several years.

This makes AMD a core supplier to OpenAI and stands as one of the largest AI chip rollouts to date.

The move comes two weeks after OpenAI's separate $100 billion equity and supply agreement with Nvidia, which set aside 10 gigawatts of capacity.

Nvidia's shares fell about 1% in premarket trading on this news.

What it means: OpenAI is diversifying away from Nvidia to avoid single-vendor dependence, which is smart infrastructure planning.

AMD gets validation that their chips can actually compete in AI workloads, which has been a question mark.

The 10% stake structure is wild—a private company potentially owning 10% of a public chip giant shows how much leverage OpenAI has right now.

The first major government report on AI job displacement just dropped, and the numbers are staggering.

Nearly 100 million US jobs could be eliminated within the next decade, according to a ChatGPT-based analysis from the Senate Health, Education, Labor, and Pensions Committee released today.

Bernie Sanders led the report.

The findings show that 89% of fast food jobs could disappear, along with 64% of accounting roles and 47% of trucking positions.

The report argues that AI's trajectory is about concentrating wealth and power by reducing labor costs and increasing corporate margins.

Republicans counter that the US must lead in AI development. They caution that heavy regulation could cede advantages to China, framing the debate as a balance between innovation and worker protections.

Democrats propose policies like a 32-hour work week, profit sharing, and a robot tax to fund worker supports and transition programs.

What it means: This is the first major government report quantifying job displacement at scale, and the numbers are massive.

If you work in fast food, accounting, or trucking, this report says your job has a high probability of automation within 10 years.

The robot tax and 32-hour work week proposals are just proposals right now—implementation would take years while job displacement is happening now.

Microsoft just rolled out two new AI agents that almost nobody can actually use yet.

Agent mode and office agent for Microsoft 365 Copilot launched this week, according to Microsoft's announcement.

The biggest limitation is availability. Office agent is only available to US-based customers with paid personal and family Copilot plans, not to most enterprise customers yet.

Agent mode is live on the web for use inside Word and Excel via the Frontier early access program. The Excel version requires the Excel Labs add-in.

The key capability people are waiting for is the Office agent inside Microsoft Copilot chat that can essentially just build PowerPoints from scratch.

This is one of the first big moves from Microsoft to bring agentic capabilities inside the normal version of Copilot that people actually use daily.

Here's what's noteworthy though y'all.

This is one of the first big public releases where Microsoft is using a model that's not OpenAI. The office agents use Anthropic's Claude Sonnet 4 and Opus 4.1 instead of GPT models.

What it means: Microsoft is hedging its AI bets by incorporating Anthropic models alongside OpenAI.

If you're a heavy Excel user, agent mode reportedly handles tasks that ChatGPT, Claude, and Gemini couldn't manage.

The Office agent's PowerPoint building capability could be a legitimate game changer once it rolls out to enterprise customers.

Sora 2 went from number one on the app store to community revolt in approximately 48 hours.

OpenAI launched Sora 2 with integrated audio and a mobile-first app that immediately hit number one.

Sora 2 produces synced dialogue, music, sound effects, and entire soundscapes while offering improved physics and realism.

The viral feature was Cameo, which lets users insert verified likenesses of themselves or their friends and create AI videos of anyone.

Sam Altman made his cameo available to everyone, and viral videos followed. Sam stealing GPUs from Target. Sam doing absolutely ridiculous things.

Then the weekend hit.

OpenAI tightened Sora's safety after people created entire South Park episodes and flooded feeds with copyrighted content.

The updated moderation filters now better catch violence, racism, and copyrighted material. But community reports say benign prompts are increasingly flagged and creative flows are disrupted.

Twitter spent the entire weekend complaining that Sora 2 got nerfed.

What it means: OpenAI launched without proper copyright guardrails and had to scramble to add them after users went wild.

If you were using Sora 2 for creative content early on, expect more restrictions as they figure out the copyright situation.

Sam Altman jokingly said "I hope Nintendo doesn't sue us" which pretty much sums up how fast and loose they launched this thing.