I attended a small event in Redmond, Washington, in early February, where Microsoft introduced the new Bing powered by ChatGPT—the first of many recent AI innovations that promise to forever change the conversation about AI. I believe it launched a transformational change in technology not seen since the introduction of the modern internet, not just with AI chatbots, but with how large language models (LLMs) will affect many aspects of our lives. Microsoft waved the proverbial green flag that day, and from that point the AI race has heated up in a way I’m not sure anyone could have predicted.
The speed and progression of announcements made since then has been staggering—and it doesn’t look like it is slowing down. The company’s latest announcement, today, the introduction of Microsoft 365 Copilot, has the potential to change the way everyday productivity tools are used by 100s of millions if not billions of people. Let's look at Microsoft's latest AI roadmap announcement and how it fits into the broader AI race.
Off to the races with Bing and Edge
Microsoft's breakthrough announcement in February introduced its new AI-powered Bing search engine and Edge browser. Despite the subsequent media circus calling out the AI’s "hallucinations" and "feelings," I think the first week of this transformational technology was relatively smooth. Microsoft said it was a work in progress; it gave every disclaimer in the book, and when it was time to reveal the findings from Bing Edge's first week, the company was humble and transparent.
I was fortunate to be among the early previewers of the new Bing and Edge. Bing has the potential to change the game for search efficiency, demonstrating that generative AI can provide natural, human-like responses within the functional context of a search engine. I have found the Bing Sidebar to be the most useful for my work, asking it to paraphrase, summarize and bulletize press releases and papers. As I have said before (and as Microsoft has acknowledged profusely), Bing still has kinks that need to be worked out. Still, the potential for this kind of contextual assistance in our daily tools could be a game changer.
Microsoft 365 Copilot improves productivity and collaboration for the future of work
Last week, Microsoft addressed a specific enterprise use case when it introduced Microsoft Dynamics 365 Copilot; this is intended to help users of the company’s CRM and ERP solution reduce repetitive tasks. That debut reinforced my hunch of where I thought the company would release its AI next: workplace productivity tools for modern work.
I discussed this on my weekly podcast here:
Now, Microsoft has gone one better with today’s introduction of Microsoft 365 Copilot. Because this will affect many different areas of Microsoft 365, including the Office apps most of us have been using for decades, this announcement has a much broader reach. The moniker is consistent with the company's Copilot nomenclature already being used across GitHub, Bing and Dynamics (since last week) the technology is not ready to stand alone; hence, a copilot.
Copilot builds on the belief that AI's current place in the workforce is to complement people in their roles rather than replace them. I believe the widespread fear that AI will take over jobs is misguided. Yet even at its current capacity, AI already has the accuracy and intelligence to automate some tasks and begin to change how people work.
Microsoft's newest Copilot is embedded into Word, Excel, PowerPoint, Teams and more, making it a companion that works alongside users in apps they already use every day. This integration makes complete sense to me; getting people to adopt new technology can be a chore, but putting a new technology next to something people already know and use—as a copilot—is a simple place to start. Earlier this month I wrote about a similar Windows 11 enhancement that uses integrated Bing Chat. Users can now quickly access Bing's AI search feature in just a few clicks from anywhere in a Windows 11 environment.
The company also announced Business Chat, a chat-based copilot experience injected into the Microsoft 365 suite. Business Chat accesses data and information across all 365 apps to quickly surface information and insights. Moreover, because it is based on an LLM, Business Chat can be programmed with organizational data, and it will continue to learn. Of course, users will have to sort and verify the information for pertinence and accuracy, the LLM nonetheless captures vital information from across the entire organization, fostering greater knowledge. I can see this saving people time searching for answers and thereby elevating their productivity, particularly in a hybrid workforce.
Just when we thought drag-and-drop low-code development was as simplified as it would get, Microsoft introduced Copilot in Power Platform. It sounds almost too good to be true, but its promise is that if you can describe something, Copilot will create it for you in a low-code experience. If Microsoft can deliver on this promise, Power Platform could become a robust tool for organizations with development shortages.
How Microsoft 365 Copilot will drive next-level efficiency
For users, an AI copilot is like having an assistant that can express itself in human, readable language. A big differentiator is that 365 Copilot harnesses organizational data and context from Microsoft Graph (more on that below). Users can also ground the 365 Copilot in a specific dataset. For example, if someone used the chatbot to write a technical brief, the AI can be constrained to capture documentation for that particular product only.
The use cases are potentially endless for how the tool will fit into daily workflows. Microsoft execs likened Copilot’s capabilities within Word to giving the word processor professional journalism skills, although I think we're still far away from that. That said, Copilot can already generate human-readable content that serves as a solid starting point for a document. That same data can then be used to populate a PowerPoint. What's incredible is how Copilot can generate a professional-looking multi-dimensional PowerPoint from pure text. I can see this as a significant timesaver, even knowing that the resultant PowerPoint deck will likely need to be reviewed and fact-checked.
These tools are nowhere near perfect yet, but Microsoft says it has already seen tremendous progress in testing. Demos during an analyst briefing showed promise of generating very informed starting points, at the very least.
Grounding the data and asking: Is this true?
365 Copilot operates with information derived from Microsoft Graph, the access point for data stored across all 365 services and products. Microsoft keeps tenant group and individual data within designated zones, meaning data is never shared or transferred between tenants—that is, between companies. Unlike most LLMs, 365 Copilot does not use data inputs to train the LLM for use by others outside the organization. This means there is no risk of unauthorized access to data between different organizations’ implementations of Copilot—and no risk that Copilot data will be used to train the LLM underlying ChatGPT. At the same time, Copilot is integrated entirely into Microsoft 365, meaning that it benefits from Microsoft’s comprehensive approach to security, compliance and privacy.
One of the biggest problems with LLMs so far is accuracy. Bing does a nice job of citing its sources so users can check the accuracy, but generative AI cannot always be trusted. Microsoft addresses this in the 365 Copilot with what it calls "grounding the query." This means that each chatbot query response is returned to the Graph to substantiate the answer. Microsoft gave the example of asking the copilot for details about a given day. The copilot would then compare the response to the calendar and other information, scanning multiple data points to answer the question: Is this true?
I've been a 365 (previously Office) user for at least 30 years, and there are features in 365 Copilot that Microsoft demoed in preview that I could see myself using on day one. That's not an easy thing to say about a technology that is transforming an industry. Other features seem a little further out; I can see people using them someday, but it will likely take a little longer for full adoption.
I like the uniqueness of the tenant group data and think that's an advantage on many levels, including security and compliance. However, Microsoft has a challenge in educating customers that the copilot is trained on an organization’s or individual's data. This means that the experience will differ significantly from Bing, ChatGPT or other generative AI bots aimed at a mass audience. With that distinction in mind, Microsoft certainly has the potential to overcome concerns that have led to ChatGPT bans in companies like Amazon, Verizon and JPMorgan Chase.
As we've seen with the introduction of the new Bing and Edge, there will definitely be mistakes and critics. As Microsoft rolls this out, people must remember that it is a copilot, not an autopilot. It won't be perfect, nor should anyone expect it to be. With that said, I look forward to seeing how Microsoft helps its customers use their data-trained LLMs to make their organizations more intelligent and efficient.
Note: Modern Work Vice President and Principal Analyst Melody Brue heavily contributed to this analysis.