In this video, the creator explores Atlas, ChatGPT’s AI browser that integrates ChatGPT and agent mode directly into the browsing experience. Atlas functions like a Chrome browser with a sidebar where users can ask ChatGPT questions and provide the current webpage as context. The agent mode allows ChatGPT to take control of the browser to perform tasks such as creating workflows, replying to YouTube comments, and even buying groceries. The creator tests the browser by asking it to find the best restaurants nearby, plan a pub crawl, and interact with various websites to see how well it performs.
The initial tests reveal some limitations, such as the browser incorrectly assuming the user’s location as London instead of their actual city, and some navigation issues with directions. However, the pub crawl planning feature impresses the creator by successfully opening a route on Google Maps. The video also includes a sponsored segment about Multilog, an anti-detect browser designed for secure multi-account management, which the creator recommends for users managing multiple AI or client accounts.
Next, the creator tests the agent mode by asking ChatGPT to draft replies to recent YouTube comments. This feature works smoothly, with the AI autonomously navigating the YouTube interface, drafting multiple replies, and impressing the creator with its ability to handle such tasks. The creator then experiments with creating an N10 workflow using the AI workflow builder, allowing ChatGPT to automate tasks based on the user’s preferences and history. Although the AI shows promise in opening the workflow builder and attempting to create workflows, some functionality still requires manual input, such as logging in.
One of the most ambitious tests involves asking ChatGPT to find a vegan recipe and add the necessary ingredients to a Tesco grocery basket. While the AI successfully identifies a recipe and attempts to add items to the basket, it struggles with login requirements and switching between supermarkets when Tesco is unavailable. The creator notes that while the AI can browse and interact with websites, it often requires user intervention, especially for tasks involving authentication. The browsing experience itself is described as somewhat clunky and less polished compared to traditional search engines like Google.
Overall, the creator concludes that while Atlas and its agent mode offer exciting possibilities for automating simple, well-defined tasks, they still fall short for more complex, open-ended human activities. The AI’s ability to navigate websites and perform actions is impressive but limited by current web design and security constraints. The creator also raises concerns about privacy and data security, given the AI’s access to browsing history and credentials. Despite these drawbacks, the concept of integrating AI directly into the browser is seen as a novel and promising step forward, with potential for future improvements as web technologies evolve.
