Microsoft has upgraded its Copilot assistant into a fully agentic AI tool that can browse the web, complete real-world tasks, and remember personal details — marking a major leap in the AI arms race.
Amazon Too
Meanwhile, Amazon has joined the push towards ‘agentic’ AI with a new shopping assistant that can browse third-party websites and make purchases on your behalf, a clear sign that AI is no longer just answering questions, but starting to take real action.
A New Era For Microsoft Copilot?
This latest overhaul appears to mark an evolution in what Microsoft calls its vision for a “true AI companion.” The trending AI news from Microsoft (on the company’s 50th birthday – yes 50!) is that Copilot now supports web-based actions, letting users issue commands that the bot then carries out independently, e.g. from booking holidays to buying gifts.
How?
Microsoft has partnered with popular platforms like Booking.com, OpenTable, Kayak, and Tripadvisor to allow Copilot to take care of these tasks without switching tabs. For example, if a user asks Copilot to “book a table for two at 7pm tonight,” it will attempt to arrange this behind the scenes.
It seems that it’s not just limited to actions. For example, the new Copilot can now do some fairly impressive, helpful, and time-saving things like:
– Track online deals and alert users to price drops.
– Analyse video in real time via the user’s phone camera to answer context-aware questions.
– View the desktop to help organise files, adjust settings, or assist with projects.
– Generate podcasts based on documents or topics users care about.
– Turn notes into structured documents using a new Pages tool.
This apparent shift towards autonomy moves Copilot closer to what Microsoft CEO of AI Mustafa Suleyman calls your ‘AI companion’. As Suleyman says: “Copilot helps you stay organised, think clearly, learn more intuitively,” and that “It’s there when you need a quick factual answer… or when you fancy just downloading after a hard day.”
Memory and Personalisation
Perhaps the most transformative (and potentially controversial) upgrade is Copilot’s memory. For example, Copilot can now learn a user’s preferences over time, e.g. favourite foods, the dog’s name, or a specific project at work. In Microsoft’s words, it remembers “not just what you said, but who you are.”
Although some may find this idea a little uncomfortable, Microsoft’s point is that this allows Copilot to offer more tailored suggestions, reminders, and proactive help. Essentially, Microsoft now wants Copilot to be a digital assistant that actually knows its user, rather than starting fresh with each interaction.
Understandably, privacy concerns are already surfacing, but Microsoft insists users retain full control. For example, users can choose what Copilot remembers, delete specific “memories,” or opt out altogether via a dedicated dashboard.
Vision, Deep Research, and More (AI That Sees and Thinks!)
The new Copilot Vision feature enables the AI to “see” what a user’s phone’s camera sees and respond accordingly. For example, a user can point their phone at a mystery plant and ask what it is, or let Copilot scan their home office and suggest layout improvements.
Meanwhile, the Deep Research function is designed for knowledge workers and business users. It pulls together information from websites, documents, and images to deliver detailed answers to complex questions, much like ChatGPT’s research capabilities or Google’s Gemini Deep Research.
Also, for anyone who’s ever spent too long scrolling, AI-generated podcasts now offer a hands-free way to absorb information. In what seems an amazing development, Copilot actually creates a back-and-forth dialogue between two synthetic hosts, based on a user’s chosen topic, with the user even able to interrupt them to ask follow-up questions.
For Business Users
While consumer features have grabbed the headlines, many of these updates actually have strong implications for business users too.
The new Pages tool lets Copilot organise a user’s scattered notes and research into a shareable, editable canvas, thereby echoing functionality found in tools like Notion or ChatGPT’s Canvas. Action-taking capabilities could save hours in admin, letting users delegate routine bookings or purchases with a single prompt.
For time-pressed professionals, Copilot can therefore become part PA, part researcher and part assistant, a proposition that could significantly enhance productivity.
Microsoft’s AI Gamble
Microsoft’s move couldn’t come at a more pivotal moment. The company has faced criticism for lagging behind OpenAI’s ChatGPT and Google’s Gemini, particularly in the consumer space. Copilot was often seen as useful but uninspiring. However, these new capabilities may be the way to place it firmly back in the race and in some areas, perhaps even a step ahead.
By integrating Copilot more deeply into Windows and expanding its personalisation, Microsoft is playing to its strengths, i.e. existing platform dominance, long-term user relationships, and enterprise credibility. It’s an ecosystem play, not just a standalone app.
However, it’s not without risk. For example, questions remain about how well Copilot can actually perform tasks on websites, or what happens when it makes a mistake, especially in business-critical settings. Microsoft hasn’t published detailed data about error rates or site compatibility and, like similar agents, Copilot can be blocked by websites concerned about revenue loss from users bypassing their apps.
Amazon’s New Shopping Agent
Not to be outdone, Amazon is testing a new agent called ‘Buy for Me’. It’s designed to go one step further, if Amazon doesn’t stock what a user’s looking for, it sends its AI to scour third-party sites, complete the purchase, and fill in your payment and shipping details. This all happens with users never having to leave the Amazon app. The AI visits the target site, selects the item, and checks out, all powered by Amazon’s Nova AI models and Anthropic’s Claude.
Possible Issues
Amazon insists this process is secure. For example, billing information is encrypted, and Amazon itself can’t see what a user is ordering from third-party sellers. However, trust is likely to be a sticking point, i.e. ensuring users feel comfortable with the idea that Amazon’s agent is able to buy the right things in the right quantity, and won’t spend too much of their money.
Also, it seems that returns and exchanges are handled via the original seller, not Amazon, so this hands-off approach could add friction if anything goes wrong.
The Rise of Agentic AI
Copilot’s updates and Amazon’s new shopping agent are part of the industry’s shift from “chatbots that talk” to “agents that do”, a trend many have dubbed agentic AI. Instead of merely suggesting actions, these tools can now perform them, often without direct user supervision.
OpenAI’s Operator, Google’s agents, and Perplexity’s prepaid-card shopping bot are all part of this wave. However, Microsoft’s scale and Amazon’s reach could see them dominate the next phase of adoption.
For users, especially professionals, this could mean offloading a growing list of digital tasks. On the downside it could also mean having to grapple with questions around accuracy, oversight, and trust. Although the app updates have arrived and AI agents can now make decisions for users, whether they should be able to and how much control users are prepared to give up are still likely to be pertinent questions.
What Does This Mean For Your Business?
By moving from reactive tools to proactive agents, Microsoft and Amazon are signalling a new era where AI doesn’t simply assist with tasks but actively takes them on. For everyday users, the promise is convenience and saved time. For businesses, particularly in the UK, it introduces new opportunities for productivity gains, automation, and customer service innovation, as well as fresh challenges around trust, oversight, and data governance.
For Microsoft, the evolution of Copilot into a more personalised and action-capable AI companion could help it stand out in a crowded market, especially among enterprise users already embedded in the Microsoft ecosystem. The ability for Copilot to act across websites, manage files, handle research, and learn user preferences may appeal to time-strapped teams juggling multiple workflows. If users can trust it and if it performs reliably, it could reduce the cognitive load and admin burden across roles, from project managers to marketing teams. However, it is whether questions around accuracy, transparency, and the limits of AI autonomy can be clearly answered that will affect adoption at scale.
Amazon’s move, while more commercially focused, appears to be pushing the envelope in a different way. Letting an AI complete purchases across third-party sites takes convenience to a new level, but also hands over a considerable amount of trust. For e-commerce operators, this could be a game-changer or a serious disruption, depending on whether their platforms are bypassed or embraced. For Amazon, it’s a chance to expand its dominance beyond its own catalogue, pulling even more of the retail experience under its umbrella.
For UK businesses, the wider implications here are worth considering. For example, AI agents could streamline internal operations, transform customer journeys, and create new service models entirely. However, they’ll also raise new questions about accountability, user control, and competitive fairness, especially as tech giants continue to roll out proprietary agents that act on users’ behalf. It seems, therefore, that the next phase of AI will not just be about what tools can say, but what they can actually do (on our behalf), and who’s actually responsible when they do it.
