"We view agents as systems that merge the intelligence of advanced AI models with access to tools, enabling them to act on your behalf and under your control," remarks Google CEO Sundar Pichai in the Google I/O announcement blog.

Project Mariner: Google’s Web-Native AI Agent

Emerging from research at Google’s DeepMind, Project Mariner is designed to act directly for the user by comprehending and interacting with web content, such as clicking, scrolling, and filling out forms as a person would. It uses advanced vision-language models to not only interpret text but also visual components like buttons, graphs, and images. This inherent multimodality—the ability to process and act on mixed media inputs—is a distinguishing feature in the expanding field of AI assistants.

The recently introduced "Teach and Repeat" functionality allows users to show an action once, like completing a complex travel form, and Mariner will replicate the sequence across similar scenarios. It can manage multiple tasks simultaneously, handling up to 10 concurrent tasks in this latest version, allowing it to independently compare and book travel options or filter data across web applications like Zillow and LinkedIn.

Agent Mode in Gemini: Intelligent Automation At Your Fingertips

Google’s Agent Mode, now active in the Gemini app for Ultra tier subscribers, transforms Mariner into a practical tool. Instruct Gemini to "find me two-bedroom apartments below $2,500 with in-unit laundry near downtown Austin," and it will conduct the search, apply filters, and display listings without any manual input. This task delegation interface merges search, automation, and conversation seamlessly. The AI not only gathers information via links but also performs digital work on your behalf.

"This is a novel and developing area, and we're enthusiastic about exploring how best to extend the advantages of agents to users and the ecosystem more extensively," stated Google on its blog.

At present, Agent Mode is available in an experimental phase for Google’s AI Ultra plan subscribers, with wider availability anticipated soon. Project Mariner is also planned to be accessible through Google's AI Mode in Search, initially via the Search Labs program.

Agentic AI Versus AI Agents: A Crucial Distinction Arises

As AI automation tools progress, so does the terminology used to define them. Experts increasingly differentiate between Agentic AI and AI Agents. Although these terms were once used synonymously, they now highlight distinct design philosophies.

AI Agents are generally purpose-driven programs meant to accomplish specific tasks, often depending on pre-written workflows, whereas Agentic AI suggests a more cognitive, self-reflective system that adapts, strategizes, and reasons dynamically without predefined steps.

Mariner aligns firmly within the Agentic AI category. It doesn't necessitate you to outline a procedure; simply express your intention, and it determines how to execute, even in unfamiliar digital settings.

This method represents a significant shift from platforms like Make.com and n8n, which empower users to create intricate AI and non-AI automation sequences provided they grasp logic trees, API endpoints, and error management. These tools, while potent, demand a level of technical expertise and initial effort. Users must manually construct each workflow: if X happens in App A, then do Y in App B.

With Mariner, the AI manages that logic instantaneously. Want to monitor invoice emails, extract totals, and update a spreadsheet? No need to sketch a flowchart. Just ask. The distinction is comparable to constructing IKEA furniture versus hiring someone who enters, interprets your thoughts, and assembles the couch before lunch.

AI-Driven Automation Gains Momentum

Similarly, conventional Robotic Process Automation (RPA) tools were crafted to simulate repetitive human actions in structured environments, such as clicking buttons, reading fields, and transferring data between legacy systems. However, they falter easily, require ongoing upkeep, and lack the contextual awareness needed for more flexible tasks. Now, RPA providers UiPath and Automation Anywhere are collaborating with Google via the Mariner API to evolve their offerings in this newest AI-powered automation era.

Conversely, Agentic AI systems like Mariner are resilient against interface modifications, can decipher new web layouts, and even resolve issues mid-process. This adaptability renders them not only stronger but fundamentally more enduring in swiftly evolving digital landscapes.

Data from Randstad Enterprise indicates that nearly 40% of RPA implementations halt due to the frailty of rule-based scripts. As AI agents grow more independent and multimodal, organizations might no longer find value in sustaining fragile automation scripts when they can merely request an AI to accomplish the task.

Competitors are swiftly entering this domain. OpenAI’s "Operator" initiative investigates analogous territories, while startups like Adept and Rewind.ai race to develop "action-taking" AI agents. What distinguishes Google is its extensive ecosystem, from Search and videos to Chrome to Gmail to Android, granting Mariner a head start.

The above is the detailed content of Google Gemini's Agent Mode And Project Mariner Shows The Future Of AI Agents. For more information, please follow other related articles on the PHP Chinese website!