One of the largest companies in the world launched computer using agents, and it may not have been a Top 5 story in AI this week.

For reals.

Such is the life in the nonstop world of AI.

Blink, and you’ve missed something huge.

(Like Microsoft releasing an autonomous AI agent that you can program to use the web or apps.)

So, what’s worth your company’s time and attention? What’s just a distraction?

Let’s dive in. 👇

OpenAI is about to drop $3 billion on AI coding company Windsurf after Cursor's parent company rejected them twice, according to CNBC.

The math is wild.

Cursor makes $200 million in revenue while Windsurf brings in just $40 million. Yet OpenAI wants the smaller fish.

Despite launching their own Codex CLI tool, OpenAI isn't waiting for organic growth. They want market share NOW.

What it means: 

OpenAI looks like they REALLY wanna snag more of the dev market. Even though they’re launching their own dev tools, it looks like they don’t wanna play this one slow.

With Google’s new Firebase Studio taking off (and for free, nonetheless), these seems more like a long-term play against Google than trying to attract attention from Claude-induced devs.

Google launched Gemini 2.5 Flash with a genius feature – adjustable "thinking budgets" for controlling AI reasoning.

The pricing? 60¢ per million tokens with reasoning off. $3.50 with it on. That's a 6X jump just for deeper thinking.

Early benchmarks show this "budget" model beating Anthropic's Claude 3.7 Sonnet while approaching OpenAI's o4 mini in reasoning tasks.

What it means: Google just fixed AI's biggest business headache – unpredictable costs.

This "pay for thinking" approach will force every competitor to copy it by fall. Companies hate paying premium rates for simple tasks, and Google delivered the solution while everyone else was busy making models that overthink everything.

Google unveiled Veo 2, their text-to-video AI powerhouse, to Gemini Advanced subscribers in stages.

It creates 8-second, 720p videos from text prompts with no monthly limits. Every clip includes Google's digital watermark but shares directly to social.

The quality gap is SHOCKING.

Veo 2 crushes competitors with realistic physics and human movement that doesn't look like glitching NPCs.

What it means: 

Google is the clear frontrunner in the AI video wars.

Veo 2 makes competitors look like toys while Google runs quantum computers. By summer, your social feeds will be flooded with AI videos so good you won't know what's real anymore.

Creator economy, meet your new overlord.

U.S. President Trump's administration is considering cutting off Chinese AI lab DeepSeek from NVIDIA chips AND banning Americans from using it, reports the New York Times.

This follows recent White House restrictions on chip sales to China, expanding on Biden-era measures.

DeepSeek gained popularity with US developers because it's cheap, forcing Silicon Valley to lower their prices.

The uncomfortable truth? Using DeepSeek means sending ALL your data to the Chinese government.

What it means: 

This isn't about tech – it's about global power with AI as the new currency.

DeepSeek's suspiciously cheap pricing was always a data harvesting operation disguised as competition.

And companies and individuals who directly used Deepseek’s API or website decided to straight up give their data to the Chinese government.

We're witnessing the start of an AI cold war that will split the global tech ecosystem by 2026.

Microsoft launched AI agents in Copilot Studio that use computers exactly like humans – clicking, typing, and navigating websites independently.

Yes. This is kinda crazy.

These agents work without APIs or integrations, dramatically expanding what businesses can automate. No coding required.

Microsoft's message is clear: if a human can use an app, their AI can too.

What it means: 

White-collar automation just got REAL. These agents will eliminate MILLIONS of administrative jobs within 18 months. No more "we can't automate that" excuses – these digital workers use the same interfaces humans do.

The office workforce transformation is happening NOW, and Microsoft beat everyone to the punch.

If you missed our recent interview with VP of AI at Microsoft, Ray Smith, this one’s worth your time.

Google announced ALL U.S. college students with .edu emails get a free year of Gemini Advanced as part of Sundar Pichai's plan to reach 500 million users by 2025.

This isn't some diet version – it's the FULL $20/month package with all premium features and 2TB of storage.

Wowzers.

Sign up by June 30, get access through spring 2026. That's nearly $400 of free AI.

Google's definition of "student" is laughably broad – anyone with a .edu email qualifies, enrolled or not.

What it means: 

Google is playing 4D chess while OpenAI fumbles with checkers.

They're trying to build an ARMY of loyal AI users who'll demand Gemini in their future workplaces.

OpenAI's (now?) measly two-month student offer looks pathetic by comparison. Can this help Google capture some users from OpenAI?

We’ll see. But right now, OpenAI is so far ahead in terms of Weekly Active Users and it’s not really close.

Anthropic rolled out Google Workspace integration for Claude, letting paid users connect Gmail, Docs, and Calendar accounts.

(Obviously…. this was hours after we kinda roasted Anthropic in this episode.)

They also launched a research tool that searches web and workplace documents – but only on their pricey Max plan ($100-200/month) or Team plans that require a minimum of five users.

Our early testing shows major problems. The Gmail integration chokes on large inboxes, often stopping after a few pages and missing emails without exact keywords.

What it means: 

Anthropic could soon be DROWNING.

They're bolting features onto Claude rather than reimagining how AI should handle personal data.

Unless they find a unique angle by summer, they'll keep bleeding users to increasingly capable and cheaper competitors. (Mainly Google on the API side and OpenAI for frontend users, especially if OpenAI acquires Windsurf.)

OpenAI is reportedly testing a Twitter-like social platform centered on their viral image generation feature.

Sam Altman is gathering private feedback, suggesting serious launch intentions. This follows X's successful Grok AI integration that competitors reportedly envy.

The strategy flips the script – instead of social networks adding AI, an AI company is adding social features.

What it means: OpenAI doesn't want social media – they want more training data.

This transparent data grab would feed directly into training future GPT/o-series models without legal complications from scraping existing platforms.

This could smoosh any first-party advantage that Grok/xAI and Meta/Llama have.

OpenAI launched GPT-4.1, a developer-focused model with a million-token context window and API prices that have competitors sweating.

This API-only model isn't on ChatGPT.com – it's built for backend development.

The pricing is AGGRESSIVE: $2 per million input tokens, $8 per million output tokens, plus 75% caching discounts.

Early benchmarks show it beating Claude 3.7 Sonnet in coding tests and GitHub reviews.

What it means: OpenAI just declared PRICE WAR on the entire AI industry.

Pew pew.

The combination of massive context and rock-bottom pricing makes GPT-4.1 irresistible for developers. Anthropic will be forced to slash rates or watch their ecosystem slowly collapse in the coming months. (Or, they’ll have to release new models aside from Claude Sonnet 3.7.)

This pricing will accelerate AI adoption by making implementation costs trivial for most businesses.

Holy frick this thing is good.

OpenAI released o3 Full and o4-Mini – models that can search, analyze images, write code, and create charts and a whole lot more all in ONE conversation.

They don't just answer questions. They THINK, reasoning through problems while autonomously using multiple tools without being told which to use.

o3 makes 20% fewer mistakes than previous versions across reasoning, coding, and visual tasks.

Users can upload whiteboard photos or diagrams and watch as the AI "thinks with" these visuals to solve complex problems.

What it means: The concept of using separate AI modes is dying.

These models represent true AI that solves problems across modalities without specific instructions. The shift from prompt engineering to collaborative problem-solving is happening NOW.

And without a doubt — this o3 model is soooooooo good. Like….. we’re still struggling to push it past its limits. (Which is usually easy for us to do.)

Should we cover o3 in more depth?

Hit us with a reply to this email if you’d wanna see/watch/read it.