Comet’s AI Assistant Evolves to Handle Complex Online Tasks

Comet’s updated assistant can interact with complex websites, operate across multiple tabs, and manage longer multi-step jobs while explicitly asking users for permission before taking browser actions. The redesign aims to reduce repetitive tab-switching for tasks like job searches, travel planning and record-keeping, and to give users clearer control over agent activity.

What changed in the assistant

Comet’s rebuilt assistant expands two practical capabilities: web perception and multi-step action. Under the product brief supplied for this story, the assistant now better parses the structure and context of complex sites and can perform a broader set of actions across browser tabs. That lets it gather, reconcile, and act on information from multiple pages without forcing the user to copy information between tabs.

The product team reported internal tests showing a roughly 23% performance improvement over the previous version on composite tasks. Rather than relying on single-step queries, the assistant is now positioned to tackle longer jobs that require ordering, persistence, and context across time.

How this affects everyday workflows

For many users, the routine online work that eats time is repetitive and cross-context: copying listings into spreadsheets, comparing fares across travel sites, or consolidating job postings from different platforms. Comet’s update targets those patterns:

• It can coordinate across tabs so a spreadsheet and several reference pages become a single working surface.

• It can parse site elements more reliably, which reduces the friction when dealing with pages that have nonstandard layouts.

• It prompts users for permission before interacting with web pages, keeping control and transparency front and center.

• “Find jobs on LinkedIn for creative software companies that are looking for product managers.”

• “Search a few major sites and find me the cheapest flight from San Francisco to New York next Monday."

• “Create a Google Sheet with a week-by-week breakdown of how often my child has been late and absent from school this year.”

Those examples map to search and curation, cross-site comparison, and structured data collection, respectively—everyday tasks in both professional and personal contexts.

Permission and user control

The redesign emphasises explicit consent. When a request would benefit from in-browser interaction, Comet asks the user whether to proceed. The user’s choice is applied for the duration of the task rather than assuming blanket access. That permission flow changes how automation feels: it places a session-level gate between the assistant’s capabilities and the user’s browser content, which addresses basic concerns around unwanted actions and surprise automation.

Developer and deployment implications

The assistant’s improved web perception creates incentives for site owners and developers to expose clearer structure and semantics so automated agents can interpret pages reliably. For organisations, this means pages built with predictable markup and accessible DOM structures may perform better when agents extract or act on content.

On the deployment side, teams will need to assess how agent interactions fit privacy, authentication and compliance requirements. The product brief flags the redesign’s intent to handle more complex jobs, but it does not disclose implementation details for credentials or sensitive data handling — areas organisations should evaluate before wider adoption.

Comet’s update targets a clear pain point: the cognitive and mechanical overhead of stitching together information across multiple web sources. For writers, researchers and product teams, the promise is fewer repetitive clicks and faster assembly of structured outputs. For developers, it signals a push toward agent-friendly web design. For privacy and security teams, it makes an explicit consent dialogue an early checklist item in assessing the technology.