🦾How You Can Leverage:

Oh, YOU didn’t launch an AI browser this week?

Awkward.

Just about everyone else did.

Aside from this week being chock-full of agentic releases, there was plenty of AI drama and billion dollar deals to be had.

What’d you miss?

(If you’re reading us daily, you’ll stay ahead. Don’t worry.)

But if life got busy and you need to tl;dr version, here’s what you need to know for AI updates this week.

How does Atlas stand out from every other agentic browser launching right now?

OpenAI unveiled Atlas this past week, a new agentic browser designed to bring your ChatGPT history into a browser. The Atlas browser is available right now for Mac users with Windows and mobile support expected soon, and free users can access it though paid accounts are required for agent mode.

The big deal? It's ChatGPT.goo

When you're browsing in Atlas and looking up competitor quarterly reports while chatting with ChatGPT on the side, it knows everything you've been working on. The browser features a ChatGPT-powered sidebar called "Ask ChatGPT" for instant web page summaries and contextual questions.

Atlas also includes a cursor collaborator feature. Anywhere you type, there's a little knob where you can bring up ChatGPT without clicking any buttons.

Perplexity made their Comet browser free for all users earlier this month, whereas previously it was only for paid users.

What it means: This is a paradigm shift in what agentic browsers can do when they pair up with your ChatGPT history. Having AI baked into a browser just makes everything faster. 2025 isn't the year of agents, it's the year of agentic browsers.

According to reports, Anthropic signed a massive cloud partnership deal with Google this past week, giving Anthropic access to up to 1 million of Google's custom TPUs in a deal reportedly valued at tens of billions of dollars.

Welp, this is Anthropic's largest TPU commitment to date.

Industry estimates put the buildout cost at around $50 billion, and it's expected to bring over a gigawatt of AI computing capacity online by next year. Anthropic's annual revenue run rate is now approaching $7 billion with Claude powering over 300,000 businesses.

Claude Code generated $500 million in annualized revenue within two months of launch. The company claims this is a record for product growth.

Here's where it gets interesting. Google has also deepened its investment in Anthropic, adding $1 billion in January to its previous $2 billion and maintaining a reported 10% equity stake.

What it means: This is an enormous deal with technically one of their competitors, but Google also has equity stake. The circular funding dynamics in AI are getting wild. This investment could help Anthropic compete better as they've been losing ground lately.

Google essentially revealed a complete rebuild of its Vibe coding section in AI Studio this past week, making it way easier for anyone to build and deploy web apps powered by Gemini without knowing how to code.

The new build tab lets users create AI-powered web apps in minutes with no upfront payment or coding skills needed.

meta Gemini 2.5 Pro is the default AI engine, but users can mix and match features like their extremely viral Nano Banana, their AI image generator Veo, and Imagine image generation. The editor now offers a split view where you can chat with AI on the left and see the full source code on the right.

Pretty much like GPT Builder inside ChatGPT.

In very Google nostalgic fashion, there's a new "I'm Feeling Lucky" button that generates random app ideas and setups with a single click.

What it means: Google is making it absurdly easy for non-technical people to build AI-powered web apps. The split view approach is familiar to anyone who's used GPT Builder. The starter templates and "I'm Feeling Lucky" button lower the barrier even more for experimentation.

OpenAI is probably going to start showing you ads pretty soon, and it's apparently Meta's fault.

(But TBH we’re kinda here for it.)

According to The Information, OpenAI is rapidly exploring personalized advertising, a shift reportedly fueled by the growing influence of hundreds of former Meta employees now working inside OpenAI. About 630 of OpenAI's 3,000 staffers are Meta veterans, including some top leadership who previously drove advertising strategy at Facebook.

OpenAI is reportedly considering using ChatGPT's new memory feature to deliver highly personalized ads based on details from users' conversations such as location and interests.

Awkwaaaard.

OpenAI CEO Sam Altman had been against this originally. But despite resistance from Altman, internal momentum reportedly driven by these Meta veterans is bringing back personalized ads and they could be coming sooner than expected.

Employees have expressed concern that OpenAI is shifting too far toward a big tech ad model.

What it means: This is happening, and it's being driven by Meta insiders inside OpenAI according to reports. Between the new Pulse feature and the Atlas browser, OpenAI is going to make a ridiculous amount of money from highly personalized ads soon.

Microsoft rolled out a major update to Copilot this past week, introducing a wide variety of features aimed at enhancing productivity, collaboration, and personalization.

The new agentic browsing features in Microsoft Edge were probably the highlight. Users now have multistep web actions like booking and unsubscribing, hands-free voice commands within Microsoft Edge, and tab reasoning to analyze content across multiple tabs.

Edge via Copilot can now autonomously complete actions on web pages.

They introduced group AI chats allowing up to 32 participants to collaborate in a single session with Copilot. The company also introduced Mico, an animated avatar that revives Clippy, and if you click it enough, there's a new "Real Talk" mode designed to get away from overly agreeable responses.

Yuuuuup. Clippy is back y'all.

Microsoft Copilot also rolled out new integration connectors to Gmail, Google Drive, Google Calendar, Outlook, and OneDrive.

What it means: The agentic browsing features in Edge are legit useful for completing multistep web tasks. Group AI chats with 32 participants could change how teams collaborate with AI. We could be getting the actual smart version of Siri from Microsoft first.

OpenAI unveiled a major new feature called Company Knowledge this past week for paid teams across the ChatGPT Business, Enterprise, and EDU plans.

The latest update allows ChatGPT to pull information directly from workplace apps like Slack, Google Drive, Gmail, GitHub, SharePoint, HubSpot, and others, making it easier for employees to find company-specific answers without switching between tools.

But there are two catches. Tmethe feature must be turned on for each new conversation, and it temporarily disables web browsing and visual output generation.

That's actually good though. Disabling web browsing helps ground it through your data instead of pulling from the internet.

This feature is powered by a reported new version of GPT-5 trained to search across multiple sources. Employees can now ask ChatGPT for project details, feedback, summaries, or performance reports and receive answers with direct links from internal company systems.

What it means: This gives employees faster, more reliable access to company data. It's similar to Gemini Enterprise or at least the grounded portion of it. Companies need to start bringing all their processes into an AI operating system now.

According to an Axios report, Meta is cutting about 600 jobs from its superintelligence lab this past week, impacting long-standing AI groups like FAIR AI Research, product-related AI, and infrastructure units.

The company's leadership believes these AI teams had become bogged down by bureaucracy.

Meta's newly formed TBD Labs is spared from the cuts according to Axios and is still actively hiring. AI chief Alexander Wang, former CEO of Scale AI which Meta acquired, said the reorganization will give remaining team members more responsibility and impact.

Let's be honest though. Meta's Llama 4 was obviously underwhelming.

There was some benchmark controversy when it first came out where allegedly they fitted one of the models specifically to score well on the LM Arena. Meta had their $14 billion-plus acquisition of Scale AI and reportedly spent more than a billion dollars on multiple individual researchers.

Not a good look when OpenAI's first open source model GPT OSS performs better than Meta's Llama 4 variants.

What it means: Meta is spending billions of dollars but simultaneously cutting hundreds of people. The expectations are very high from whatever Llama ships next. There's a ton of pressure on Alexander Wang and the new TBD Lab to deliver actual results.

According to reports from The Verge, Microsoft is preparing a major transformation of its Outlook email client, aiming to essentially rebuild it from the ground up with AI at the core rather than sprinkling AI on top.

Holllllld up.

The rebuild is described as the future Outlook as a "body double" that will help users tackle work, making tasks feel less overwhelming by acting as a true partner, not just a set of tools.

Copilot will play a central role in this transformation, turning Outlook into a proactive partner that reads messages, drafts replies, and organizes schedules automatically. Essentially, it will read your emails, prioritize what's important, draft replies, and organize your schedule automatically according to what's going on in your Outlook inbox.

The Outlook team is shifting to rapid development with weekly feature experiments and prototyping cycles measured in days instead of months.

What it means: Microsoft is rebuilding Outlook from the ground up with AI at the core. The "body double" concept means Outlook will act as a true partner that tackles work for you, not just another productivity tool you manage.

Here's the thing about OpenAI reportedly building a new AI music tool. It could actually be pretty significant.

According to a recent report in The Information, OpenAI is reportedly building a new AI-powered music generation tool that creates music from just text and audio prompts. The tool aims to allow users to add custom music to videos or generate instrumental tracks, such as guitar accompaniment for vocals.

OpenAI is collaborating with Juilliard School students to annotate musical scores, providing high-quality training data for the model.

What people don't really know is OpenAI was actually one of the first platforms to have an AI music platform called Jukebox way back when, but no one really used it. It remains unclear whether this new tool will be a standalone product or just integrated into existing OpenAI apps like ChatGPT or Sora.

Nah, this isn't a super crowded space. Udio and Suno are pretty much running away with it.

What it means: This should be a fun battle to watch between OpenAI, Suno, and Udio. Sora 2's music quality was surprisingly good, which suggests OpenAI may have already been testing this technology. The Juilliard collaboration shows they're taking music quality seriously.