ChatGPT Agents and Google’s Mariner: The Rise of Multimodal AI in July 2025

Explore ChatGPT Agents & Google's Mariner in July 2025, driving the multimodal AI revolution with web automation and proactive task execution.

July 29, 2025
9 min read

Introduction: A New Era of AI Interaction

Imagine a world where your digital assistant doesn’t just answer questions but acts on your behalf—booking flights, filling out forms, or even researching complex topics while you sip your morning coffee. It’s not science fiction anymore. By July 2025, the AI landscape has transformed dramatically, with ChatGPT Agents and Google’s Project Mariner leading the charge in what’s being called the “agentic era” of multimodal AI. These aren’t your average chatbots; they’re intelligent, action-oriented systems that blend text, vision, and real-world execution to redefine how we interact with technology.

But what makes these AI agents so revolutionary? Why is July 2025 a pivotal moment for multimodal AI? And how do ChatGPT Agents and Google’s Mariner stack up in this brave new world? Let’s dive into the story of how these technologies are reshaping our digital lives, backed by the latest research, real-world applications, and expert insights.

What Is Multimodal AI, and Why Does It Matter?

Multimodal AI is like a Swiss Army knife for artificial intelligence. Unlike traditional AI, which might focus solely on text or images, multimodal systems process and generate multiple data types—text, images, audio, video, and even web interactions—seamlessly. Think of it as an AI that can see, read, talk, and act like a human assistant, all at once.

In July 2025, multimodal AI is no longer a niche concept—it’s a game-changer. According to a report from Indigo.ai, the global AI agent market, valued at $5.4 billion in 2024, is projected to skyrocket to $47.1 billion by 2030, with a compound annual growth rate of 44.8%. Why the boom? Multimodal AI agents like ChatGPT Agents and Google’s Mariner are moving beyond answering questions to executing tasks, making them indispensable for businesses, developers, and everyday users.

The Agentic Era: From Passive to Proactive AI

The term “agentic AI” is buzzing in tech circles, and for good reason. Unlike earlier AI models that waited for your input, agentic AI systems like Mariner and ChatGPT Agents are proactive. They don’t just respond—they plan, reason, and act. As Google’s CEO Sundar Pichai noted, “If Gemini 1.0 was about organizing and understanding information, Gemini 2.0 is about making it much more useful”. This shift from passive to proactive AI is the heart of the multimodal revolution, and July 2025 marks a turning point.

Google’s Project Mariner: The Web-Surfing AI Agent

Picture this: you’re planning a weekend getaway, but the thought of navigating travel websites feels like a chore. Enter Project Mariner, Google DeepMind’s experimental AI agent that surfs the web for you. Unveiled in December 2024 and expanded at Google I/O 2025, Mariner is a Chrome extension powered by Gemini 2.0, Google’s most advanced multimodal AI model to date.

How Does Mariner Work?

Mariner operates on an Observe–Plan–Act loop, mimicking human web navigation:

Observe: It “sees” the browser, analyzing text, images, forms, and even the DOM structure of a webpage.
Plan: It reasons through your request, creating a step-by-step action plan.
Act: It executes tasks like clicking buttons, filling forms, or extracting data, all while showing you its process for transparency.

For example, in a demo shared by TechCrunch, a user asked Mariner to “add all the veggies from this recipe to my Safeway cart.” Mariner navigated to Safeway’s website, searched for items, and added them to the cart, checking off each step as it went. It’s like having a digital assistant that not only understands your grocery list but also does the shopping for you.

Key Features of Project Mariner

Multimodal Reasoning: Mariner processes text, images, and web layouts, making it adept at handling complex websites.
Teach and Repeat: Show Mariner how to complete a task once, and it can replicate it later, perfect for repetitive workflows like filling out travel forms.
Parallel Task Handling: It can manage up to 10 tasks simultaneously in cloud-based virtual machines, freeing you to work on other things.
Deep Google Integration: Mariner ties into Google’s ecosystem, including Search, Gmail, and Drive, for personalized experiences.

Limitations and Challenges

Mariner isn’t perfect yet. As of July 2025, it’s still in an experimental phase, available only to Google’s AI Ultra subscribers ($249.99/month) and select developers. It’s also slow—sometimes taking seconds between actions—and can’t handle sensitive tasks like entering payment details. Early testers have noted that Mariner requires an active browser tab, limiting multitasking. Google acknowledges these issues, with Labs Director Jaclyn Konzelmann stating, “It’s not always accurate and slow to complete tasks today, but it will improve rapidly over time”.

ChatGPT Agents: OpenAI’s Answer to Agentic AI

On the other side of the ring, OpenAI’s ChatGPT Agents, powered by the Computer Using Agent (CUA) model (a variant of GPT-4o), are making waves. Announced in July 2025 alongside OpenAI’s AI-powered web browser, these agents are designed to “delegate and execute” rather than just search and respond.

How ChatGPT Agents Work

ChatGPT Agents operate similarly to Mariner, using a cloud-based execution model to navigate websites and perform tasks. Built on GPT-4o, they combine vision, text, and reinforcement learning to interact with graphical interfaces without needing custom APIs. For instance, you could ask a ChatGPT Agent to “book a table for two at a Michelin-starred restaurant in San Francisco,” and it would navigate reservation platforms, fill out forms, and confirm the booking—all while asking for clarification if needed.

Key Features of ChatGPT Agents

Action-Oriented Design: Agents can click, type, and navigate websites, turning any webpage into an actionable interface.
Contextual Memory: With OpenAI’s recent “memory” feature, agents retain user preferences for personalized task execution.
Multimodal Inputs: They handle text, images, and documents, making them versatile for tasks like analyzing spreadsheets or summarizing web content.
Enterprise Applications: ChatGPT Enterprise integrates agents into workflows for content creation, customer support, and data analysis.

Limitations and Challenges

Like Mariner, ChatGPT Agents are still evolving. They struggle with tasks requiring logins, payments, or CAPTCHAs, often needing human intervention. Testing reveals occasional “hallucination” issues, such as incorrect dates or booking errors, which can undermine reliability for critical tasks. OpenAI is addressing these through iterative updates, but as of July 2025, the agents are not yet foolproof.

Head-to-Head: Mariner vs. ChatGPT Agents

Both Mariner and ChatGPT Agents are pushing the boundaries of multimodal AI, but they have distinct strengths and weaknesses. Here’s a quick comparison based on the latest insights:

Feature	Google’s Project Mariner	ChatGPT Agents
Underlying Model	Gemini 2.0	GPT-4o (CUA)
Web Navigation	Chrome extension, cloud-based VMs	Chromium-based browser, cloud execution
Task Capacity	Up to 10 simultaneous tasks	Limited parallel tasks (exact number unclear)
Ecosystem Integration	Deep ties to Google Search, Gmail, Drive	Limited to OpenAI’s platform, some third-party integrations
Performance	83.5% on WebVoyager benchmark	38.1% on OSWorld benchmark
Availability	AI Ultra subscribers, select developers	Limited beta, broader release expected soon
Strengths	Multimodal reasoning, teach-and-repeat feature	Strong conversational abilities, memory feature
Weaknesses	Slow, active tab requirement	Hallucination issues, login/payment limitations

Sources:,,

Which Is Better?

It depends on your needs. Mariner excels in web-based automation and benefits from Google’s vast ecosystem, making it ideal for users embedded in Google services. ChatGPT Agents shine in conversational flexibility and enterprise applications, leveraging OpenAI’s strength in natural language processing. As Forbes noted, Mariner’s “native multimodality” gives it an edge in handling visual web elements, while OpenAI’s focus on “unified intelligence” makes its agents more adaptable for creative tasks.

Real-World Impact: How Multimodal AI Is Changing Lives

The rise of multimodal AI agents isn’t just a tech trend—it’s transforming industries and daily life. Here are some real-world examples:

E-Commerce: Mariner can compare prices across retailers and add items to carts, while ChatGPT Agents optimize customer journeys by automating purchases.
Productivity: Businesses are using ChatGPT Enterprise to automate content creation and customer support, with agents drafting emails or summarizing reports.
Research: Mariner’s “Deep Research” feature generates comprehensive reports by scouring the web, rivaling OpenAI’s o1 model for multistep reasoning.
Accessibility: Google’s Project Astra, which shares roots with Mariner, narrates live camera feeds for blind users, showcasing multimodal AI’s potential for inclusivity.

A case study from HackerNoon highlights how Mariner automates webinar registrations by filling out forms and selecting preferences, saving users hours of repetitive work. Meanwhile, OpenAI’s enterprise clients, like those in pharmaceuticals, use ChatGPT Agents to streamline data analysis and compliance tasks.

Ethical Considerations: The Double-Edged Sword

With great power comes great responsibility. Multimodal AI agents raise ethical questions:

Privacy: Both Mariner and ChatGPT Agents access sensitive web data. Google emphasizes “scoped permissions” to protect user data, while OpenAI uses cloud-based execution for centralized safety controls.
Job Displacement: Automation could disrupt roles in customer support or data entry. Mitigation strategies include upskilling workers for AI-related jobs.
Bias and Misinformation: AI agents can amplify biases or generate false information. Google and OpenAI are investing in rigorous testing and explainable AI to address this.

As Adyog notes, “Transparent data practices and user control are critical to building trust in AI agents”. Both companies are under pressure to balance innovation with ethical responsibility, especially as regulators scrutinize Google’s dominance and OpenAI’s monetization strategies.

The Future of Multimodal AI: What’s Next?

July 2025 is just the beginning. OpenAI’s GPT-5, expected in summer 2025, promises “unified intelligence” with deeper contextual memory and advanced agent capabilities. Google is expanding Mariner’s reach through the Gemini API, aiming for broader developer access by late 2025. Meanwhile, competitors like Anthropic’s Claude and Amazon’s Nova Act are entering the agentic AI race, signaling a fragmented but exciting future.

The web itself is changing. As Azoma predicts, “The browser becomes an execution environment rather than an information discovery tool”. Brands will need to optimize for “agent discovery” rather than traditional SEO, ensuring websites are AI-friendly with clear structures and predictable flows.

How to Get Started with Multimodal AI Agents

Ready to dive in? Here’s how to explore ChatGPT Agents and Project Mariner:

For Mariner:
- Join the trusted tester waitlist on the Project Mariner website.
- Experiment with demo workflows like job listing searches or email triage.
- Use precise prompts, e.g., “Book a flight to Paris under $500 for next weekend”.
For ChatGPT Agents:
- Sign up for ChatGPT Enterprise or wait for the public release of OpenAI’s AI browser.
- Test tasks like drafting emails or analyzing documents, leveraging the memory feature for personalization.
- Provide iterative feedback to refine agent performance.

Conclusion: The Dawn of a New Digital Frontier

In July 2025, ChatGPT Agents and Google’s Project Mariner are more than just tech innovations—they’re harbingers of a new way to interact with the digital world. From automating mundane tasks to empowering businesses and individuals, multimodal AI is turning the internet into a playground for proactive, intelligent agents. As Google’s Jaclyn Konzelmann put it, this is a “fundamentally new UX paradigm shift”. Whether you’re a developer, a business owner, or just someone tired of filling out forms, these AI agents are here to make your life easier.

So, what’s your next step? Will you let Mariner surf the web for you or task a ChatGPT Agent with your next big project? The future is here, and it’s multimodal, proactive, and ready to act. Let’s embrace it.

Sources: Cited throughout the blog with references.