• Everyday AI
  • Posts
  • Agentic AI in the Browser: The next frontier of artificial intelligence?

Agentic AI in the Browser: The next frontier of artificial intelligence?

OpenAI and UK government join forces, Google eyes news licensing for AI content, OpenAI and Oracle boost power for Project Stargate and more!

Author

Everyday AI
July 22, 2025

Outsmart The Future

Today in Everyday AI
6 minute read

🎙 Daily Podcast Episode: Perplexity, OpenAI, and Microsoft are going all-in on agentic AI in the browser—leaving chatbots behind. Discover why agentic browsers are the next frontier in artificial intelligence. Give it a listen.

🕵️‍♂️ Fresh Finds: ChatGPT gets over 2.5B prompts daily, Netflix is using Runway’s AI video tools and Universal Music Group increases AI music patent efforts. Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: OpenAI and UK government join forces, Google eyes news licensing for AI content and OpenAI and Oracle boost AI data power for Project Stargate. For that and more, read on for Byte Sized News.

🧠 Learn & Leveraging AI: Are agentic AI browsers the new chatbots? We break down the possibilities and what it means for the AI world. Keep reading for that!

↩️ Don’t miss out: Did you miss our last newsletter? We talked about OpenAI making Instacart CEO head of product growth, Meta’s $1.25B job offer, Grok 4 boosting xAI’s revenue over 300% and more. Check it here!

 Agentic AI in the Browser: The next frontier of artificial intelligence? 🤔

LLMs are so yesteryear. The next wave?

Agentic browsers.

While we're all rushing to bring personalization, company files and more into front-end large language models, agentic browsers have been quietly staking their claim as the next big thing in AI.

Also on the pod today:

• Big Tech Agentic AI Browsers 🏢
• Agentic AI Browsers vs. Chatbots 💭
• Advantages of Agentic AI Browsers 🧑‍💻️

It’ll be worth your 34 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – Inferless helps you deploy Machine Learning models in minutes, Podcastle launched a suite of high quality AI voices, Jotform's new AI Agents organize requests to help you get stuff done.

OpenAI – OpenAI says ChatGPT users send over 2.5 billion prompts every day.

ChatGPT’s new personality feature is showing up for some on the iOS app.

OpenAI’s new CEO of Applications released an article on her future role and goals.

AI in Media – Netflix is using Runway’s AI video tools for production.

AI Research – Researchers from top AI labs like OpenAI and Google are warning they may be losing the ability to understand advanced AI.

AI Music - Universal Music Group is increasing its efforts on music AI patents.

AI in Society – 72% of U.S. teens have used AI companions, a study finds.

1. OpenAI and UK Government Join Forces to Boost AI Growth 🇬🇧

In a timely move to advance the UK’s AI ambitions, OpenAI and the UK Government have signed a Memorandum of Understanding to accelerate AI adoption across public services and private sectors. This strategic partnership aims to create high-skilled tech jobs, improve government efficiency, and support small businesses with AI-powered tools, building on the UK’s AI Opportunities Action Plan.

With OpenAI already powering key government chatbots and expanding its London office, this deal signals a major push to embed AI deeply into Britain’s economy and infrastructure.

2. Google Eyes News Licensing for AI Content Boost 📰️

Google is launching a pilot licensing program with about 20 national news organizations to source content for its AI projects, signaling a shift toward paying publishers amid rising industry tensions. This move comes as startups like OpenAI and Perplexity AI have already started compensating media companies, offering them critical revenue streams.

Google’s project aims to tailor licenses to specific AI products, potentially reshaping how digital news integrates with AI-generated summaries that currently siphon web traffic.

3. OpenAI and Oracle Boost AI Data Center Power in U.S. ⚡️

OpenAI and Oracle have announced plans to add 4.5 gigawatts of new data center capacity to their Project Stargate initiative, pushing the total to over 5 gigawatts in an effort to maintain U.S. leadership in AI infrastructure, Reuters reports.

This expansion is part of a massive $500 billion investment with partners including SoftBank, aimed at supporting AI services like ChatGPT that demand immense computing power. While specific locations and funding details remain under wraps, the move signals a serious push amid global AI competition and national security concerns.

4. Microsoft Poaches AI Talent from Google DeepMind 👀️

In a rapid talent grab, Microsoft has hired around two dozen top AI researchers and engineers from Google DeepMind, signaling a heated battle for AI expertise among tech giants. Notable departures include Amar Subramanya and Adam Sadovsky, both now corporate VPs at Microsoft AI under CEO Mustafa Suleyman, a DeepMind co-founder who joined Microsoft last year.

This move comes as Microsoft doubles down on its AI ambitions with products like Copilot, even while trimming its workforce by 9,000 employees.

5. xAI Seeks Perpetual Access to Employee Face Data for AI Training 🫥

In a recent move reported by Business Insider, xAI asked its staff in April to record their facial expressions to help train Grok, its AI designed to understand human emotions. Employees reportedly signed forms granting the company ongoing rights to use their likeness for training and commercial purposes.

This raises timely questions about privacy and consent as companies increasingly rely on biometric data to advance AI capabilities.

6. Apple’s AI Team Faces Turmoil Over Open-Source Plans 😬

A recent report from The Information reveals internal conflict at Apple’s AI division, where a push to open-source key AI models was blocked by software chief Craig Federighi, citing concerns over performance compromises on iPhones. The on-device-first strategy, prioritizing user privacy, is reportedly limiting AI capabilities and prompting top talent to leave.

Meanwhile, Apple is reconsidering compensation to retain staff and exploring partnerships with OpenAI, Anthropic, and Google for Siri’s AI upgrade.

🦾How You Can Leverage:

Something just outpaced LLM development speed.

Impossible?

Welp… it’s happening. 

We've tracked AI daily for almost three years. Nothing moves faster than large language model innovation. Ever.

Here's the prediction: Perplexity's Comet browser will generate more momentum than Perplexity's chatbot in 2025. Same company. Different future.

The agentic browser movement got the Everyday AI treatment today because every major tech company dropped agentic browser updates simultaneously. Microsoft Edge. OpenAI's agent. Google's Project Mariner.

Wild timing.

You're about to discover why front-end chatbots create unnecessary separation from actual work. How "teach a task mode" eliminates repetitive workflows forever. And why power users can save hours daily by switching approaches.

Let’s get to the three big takeaways from today’s show.

1 – The Middle Man Gets Eliminated 🔥

Front-end chatbots feel like unneeded separation between your data and your goals.

Wrong.

Every time you upload files to ChatGPT, explain context, then copy-paste results back? That's three extra steps killing productivity. Knowledge workers have these automated processes ingrained in their brains.

Agentic browsers cut out the middleman completely. Your tabs become memory. Your login state becomes instant context. The traditional context window disappears because everything happens where you're already working.

Consider this reality: You need data from ten sites daily. Buzzsprout. Beehive. Google Analytics. Spotify. Apple podcasts. YouTube.

Traditional method? Download 10 CSV files. Clean data. Upload to LLM. Cross fingers.

Agentic browsers access everything you're logged into. Dynamic data without downloads. No format gymnastics.

Try This:

Count every website you log into daily for work.

Now imagine an AI accessing all simultaneously while you stay logged in. Fellou AI's shadow windows already automate cross-site tasks. Opera Neon books travel autonomously.

Test with your three most-visited work sites first.

2 – Speed vs Power Get Solved 🔥

OpenAI’s Operator is powerful but extremely slow. Agentic browsers solve this with hybrid architecture.

Here's what we discovered testing approaches.

OpenAI's operator uses computer vision and screenshots to navigate. Impressive capabilities.

Painfully slow execution.

Agentic browsers bring processing on-device instead. Like NPU AI chips in smartphones, basic operations stay fast without constant server roundtrips.

Google's Project Mariner revealed the game-changer: teach a task mode.

You perform actions once. It learns the pattern and repeats indefinitely.

Zero configuration required.

Try This:

Pick one repetitive task spanning multiple websites.

Walk through it manually while noting each step. Document the clicks, navigation, data points. That's exactly what these browsers learn and automate.

Start with something low-stakes like social media analytics across platforms.

3 – Less Duct Tape Philosophy Wins 🔥

MCP servers require technical setup. 

This matters more than tech blogs admit.

Model Context Protocol revolutionized AI communication and Agent 2 Agent protocols create seamless data sharing. Both demand server configuration and monitoring.

Brilliant for developers. Overkill for executives looking for an easy leg up. 

We've seen MCP explosions recently and also surges of problematic MCPs and phishing scams. Bad actors capitalize on rapid adoption.

Agentic browsers skip that complexity by keeping your data connections local. 

The Dia browser routes queries through skills systems selecting the best LLM per task. Local browser cookies provide safe contextual assistance. No custom servers. No protocol configuration.

You login once, work normally and let AI observe and assist.

The trade-off, though? Slower than perfect MCP setups. More stable than protocol failures and security vulnerabilities.

For most business applications, simplicity wins.

Try This:

List every tool and workaround you use getting data into LLMs.

Count steps between questions and useful answers. Now imagine cutting that by 80% through direct browser access.

Download Perplexity's Comet and test one cross-site task.

Reply

or to participate.