In celebration of its 50th anniversary, Microsoft has enhanced its AI chatbot, Copilot, by introducing new capabilities.
According to Microsoft, Copilot can now execute actions on the majority of websites, facilitating tasks like booking tickets, making restaurant reservations, and other similar activities. Additionally, the chatbot can retain specific personal preferences about users, akin to functionality seen in OpenAI‘s ChatGPT, such as favorite foods and movies. Copilot has also gained the ability to interpret real-time video captured by your phone, responding to queries concerning what it visually identifies.
These advancements come at a time when Microsoft is reportedly reconsidering the structure of Copilot, traditionally reliant on OpenAI‘s AI models, potentially shifting more towards its own proprietary technology. Historically, Copilot has trailed behind competitors like ChatGPT and Google‘s Gemini, both of which have significantly accelerated the introduction of new features recently.
Starting Friday, Copilot will be capable of handling web-based tasks comparable to “agentic” assistant tools such as OpenAI‘s Operator. Microsoft has established initial integrations with partners including 1-800-Flowers.com, Booking.com, Expedia, Kayak, OpenTable, Priceline, Tripadvisor, Skyscanner, Viator, and Vrbo, allowing immediate compatibility. Users simply input commands—for instance, “order flowers for my spouse”—and Copilot will strive to complete the requested action.
Borrowing ideas from search-focused tools like Perplexity, Copilot now possesses the ability to monitor online sales and discounts on your behalf. Instructing the platform to watch for specific price reductions or promotions means that it will inform you when opportunities arise, offering direct purchase links as well.
How effectively Copilot can conduct these various tasks remains uncertain, as Microsoft has provided limited information about the inner workings of this new functionality. Unlike several rival systems, Microsoft has not released specific performance metrics indicating scenarios where Copilot may encounter difficulties or necessitate human assistance.
It’s plausible that certain websites may opt to block Copilot, similar to their ability to restrict OpenAI‘s Operator access. A website might take such measures to prevent potential declines in direct traffic and associated ad revenues.
The positive aspect is that Copilot‘s other new additions are clearer in purpose and likely carry fewer concerns regarding controversy or uncertainty.
The refined Copilot can now produce synthetic “podcasts” similar in concept to the Audio Overviews available in Google‘s NotebookLM. Given a webpage, document, or other informational content, the chatbot will create a conversational exchange featuring virtual hosts. Users can interrupt this discussion at any moment to pose specific questions, prompting responsive engagement from the artificial hosts.
Moreover, on Android and iOS devices, Copilot is now capable of processing what appears in your smartphone’s camera or image gallery. Users may query about visible objects—for instance, “what type of flower is this?” On Windows, the updated Copilot application gains the functionality to actively view your desktop screen content, enabling it to conduct searches, modify settings, manage and structure files, and other desktop-related tasks. Initially, this functionality will become accessible to those enrolled in the Windows Insider program starting next week.
This author hopes that suitable security measures exist to ensure privacy by preventing Copilot from accessing confidential files or inadvertently causing desktop disruptions. However, details on such safeguards were sparse at the time of reporting.
Furthermore, Copilot introduces a new feature called “Pages,” designed for compiling projects and heavily influenced by existing tools like ChatGPT‘s Canvas and Anthropic‘s Claude Artifacts. Pages serves as an interactive workspace, consolidating research and notes, with Copilot simplifying organization and content creation processes.
In conjunction with Pages, a new capability called Deep Research enables Copilot to extract, evaluate, and synthesize information from sources such as online articles, documents, and images, addressing more challenging inquiries, comparable to the Deep Research features previously implemented by ChatGPT and Google‘s Gemini.
Lastly, Microsoft enhanced Copilot‘s capacity to retain more individualized user data about previous interactions. The company indicated Copilot will learn and recall your personal preferences over time, proposing customized guidance, anticipatory recommendations, and timely reminders based on previous interactions.
For users concerned about privacy implications of a chatbot tracking their personalized information, Microsoft emphasizes the availability of privacy controls. Users have the freedom to selectively remove saved “memories” or entirely disable this personalization feature.
Copilot puts users in control through its dashboard, offering options to specify which personal information it keeps or to opt out of data collection altogether. Your control remains firmly in your hands.