Perplexity AI has introduced a significant upgrade to the Comet Assistant, with CEO Aravind Srinivas highlighting a renewed emphasis on transparency, user choice, and safer automation for online tasks. Announced on Saturday, this update aims to provide users with a clearer insight into the workings of the browser-based AI agent while ensuring its operations are limited to user-approved permissions. Srinivas noted that the revamped Comet has been designed to function more like a responsible human assistant—being visible, accountable, and cautious about when to act independently. The company expects these enhancements will boost users’ confidence in allowing the assistant to manage complex online activities. Comet now provides a step-by-step display of its actions.
In a comprehensive blog post, Perplexity detailed that users will have a direct view of everything the assistant does while interacting with a webpage, including browsing, clicking links, scrolling, or navigating through multi-step processes. A new sidecar panel adds clarity by showing the assistant’s reasoning and enabling users to track its workflow in real time. This approach, according to Perplexity, ensures that users are always informed about why Comet takes specific actions and how it arrives at its conclusions. Another key enhancement relates to how and when Comet assumes control of tasks. When a query is entered in the omnibar or search interface, the assistant may inquire if it should perform the task independently.
Users are presented with multiple choices—managing the task themselves, permitting Comet to execute the action once, or providing ongoing permission. Perplexity underscores that this system prevents the assistant from overstepping boundaries, allowing users to automate tasks only when they feel at ease. Acknowledging that certain actions necessitate additional caution, the updated Comet Assistant will pause when encountering high-risk steps, such as logging into accounts or processing purchases. In these scenarios, the assistant explicitly requests permission before proceeding. Perplexity states that this reflects the judgment of a trustworthy human aide—someone who recognizes when a decision is too critical to make without confirmation.
These updates to Comet align with Perplexity’s broader objective of developing reliable, transparent, and user-centric AI tools across its ecosystem, which includes Deep Research, Labs, Email Assistant, and Background Assistants. The company aims to shift the user perspective from questioning what an AI can do to determining what they want the assistant to achieve, all while being assured of its safe and transparent operation.
