Better AI Functionality in Microsoft 365 Apps and Agents Everywhere
On September 16, Microsoft launched Microsoft 365 Copilot Wave 2 (note the name change from Copilot for Microsoft 365). Wave 2 is a broad refresh of Copilot capabilities to achieve several goals, including:
- Make Copilot much more of a reality for applications like Excel and PowerPoint where the previous implementation was not good. Excel, an application where 90% of its available features are often ignored by users, gets turbocharged analysis through Python. According to Microsoft, Python is “one of the world’s most popular programming languages for working with data.” I don’t doubt this statement for an instant, but I do wonder whether Python will affect the way most users interact with worksheets. Because it helps people build presentations, the new Narrative Builder could be a big hit, especially as it can use corporate branding.
- Introduce the Prioritize my Inbox feature for the new Outlook for Windows (not Outlook classic). I can’t help but think that this is version 3 of Microsoft’s attempt to help people make sense of a crowded Inbox. The previous versions (Clutter and the Focused Inbox) aren’t as sophisticated when it comes to the technology deployed to analyze the inbox. There’s no doubt that Copilot does a better job of highlighting important emails, but I’m less enthusiastic about inbox prioritization than Microsoft is. Maybe it’s because I’ve spent too much time attempting to train Clutter and the Focused Inbox about what email is important to me.
- Launch Copilot Pages, a Loop component to capture the results generated with Biz Chat (Business Chat), the new name for Microsoft 365 Copilot chat (grounded in the Graph). Microsoft calls Copilot Pages “the first new digital artifact of the AI age,” whatever that means. On a practical level, it makes perfect sense to capture Copilot interaction in a form that is easily shared and collaborated with other users. It will be interesting to see how users take to this idea.
- Deliver Copilot for OneDrive with features like the ability to compare and summarize documents. OneDrive is a personal repository and people really shouldn’t store mission-critical information there because of the issues that can occur when people leave the organization, including how to deal with PII and other personal data. But humans are humans, and people do keep valuable data in OneDrive. It’s good to have some new features to make more sense of what’s kept there. Perhaps an agent will come along to help people move identity and move important files from OneDrive to suitable SharePoint sites.
- Make information more accessible through agents, personal AI assistants focused on specific sets of knowledge. Microsoft has big plans for agents. I think this is the biggest and technically the most interesting advance, especially in how agents leverage SharePoint Online content. We discuss agents below.
There are a bunch of other changes too, like Teams chat now being incorporated into meeting recaps and Word being able to reference emails and meetings in addition to documents. These changes reflect the experience of using Copilot over the last year.
Overall, Wave 2 signals a move from the initial introduction of generative AI technology across all the Microsoft 365 apps to making it easier for people to consume and exploit the technology. Along with improvements in performance, stability, and quality, Wave 2 of Copilot is all about helping customers better exploit the information they can access through the Graph in Teams, SharePoint Online, OneDrive for Business, and Exchange Online.
A set of videos are available online to help you understand how these changes operate. Remember that what you see in the videos represents what Microsoft plans to deliver. Changes can happen between creating a video and software delivery, but the videos are useful in terms of understanding where Microsoft is heading,
A Profusion of Agents
Copilot agents in SharePoint will be available in preview in early October. These are lightweight versions of the agents constructed with Copilot Studio, which allows developers to build agents to interrogate repositories such as web sites. One of the issues of using generative AI to query bodies of knowledge like the internet is that so much of what might be found and reused is inaccurate, obsolete, or misleading. Even within Microsoft 365, Graph-based queries can return inaccurate or obsolete information or reveal items that users shouldn’t see. This is why Microsoft introduced Restricted SharePoint Search to limit Copilot prompts to information from a curated set of sites.
When you limit queries to curated sets of knowledge, it’s inevitable that better quality responses can be generated. Copilot agents leverage this truth by using curated data selected from a SharePoint Online document library or folder by the person who creates the agent. As an example, I could build an agent to query the articles I’ve written about Microsoft Copilot 365, all of which are stored in a single folder in a SharePoint Online site. Interacting with the agent should then deliver much more precise information than a general trawl to find anything relevant in the Graph. Microsoft’s demos focused special attention on how support teams could leverage SharePoint agents in this way (Figure 1).
In addition, Microsoft says that every SharePoint Online site will have a prebuilt Copilot agent that’s grounded in the information contained in the site. The idea here is that the prebuilt agent will allow users easier access to documents stored in the site. Given that people add 2.5 billion new files to SharePoint Online daily, it’s a good idea to help users distinguish the wood from the trees. Although I can’t confirm this yet, it seems like data from SharePoint pages and Microsoft Lists cannot be included in the scoped set for an agent.
Agent configurations are stored as JSON files with a copilot extension. These files can be managed just like any other SharePoint file, including support for retention labels and sensitivity labels. Because the agent is represented as a file, it can be added to a Teams chat or shared in email to allow others to interact with the agent, providing that the other users have access to the set of scoped documents. The ability to converse with the agent in Teams chat through @ mentions is a nice touch.
If desired, an agent can be further enhanced in Copilot Studio, such as including information sources outside SharePoint Online.
No New Access – Just a New Way to Engage with Copilot
Some might fear that people will misuse agents and be able to use agents to find information that is otherwise inaccessible. Microsoft’s view is that this cannot happen because everything an agent can retrieve is already available to a signed-in user with a Microsoft 365 Copilot license.
The same access controls that apply to Copilot interactions today apply to the prompts processed by Copilot agents in SharePoint. People using agents to retrieve information cannot see the content contained in the target items unless they have permission to do so. Any restrictions in place, such as the sensitivity label setting to stop Copilot sending documents to Microsoft content services for processing, are respected. Any other restriction currently applicable to using Copilot with SharePoint, such as a 3 MB maximum file size, is also valid.
Customer IT Departments Might Not Like Agents
The view from corporate IT departments about the arrival of new Copilot features might be less sanguine than Microsoft expects. Already under pressure to understand and manage the introduction of AI technology into the business alongside other projects, IT departments might look at the ability of users to create Copilot agents with some dread.
Rolling out any new technology takes preparation. Introducing Copilot agents to the average user, even those who have been using Microsoft 365 Copilot for several months, is probably not a good idea if it’s not backed up with support and user training. And the work to rein in poorly managed SharePoint sites and reduce oversharing and digital rot to an acceptable level might not yet be complete. Adding agents to the mix seems like a stretch.
The fact that Microsoft plans to introduce Copilot agents in SharePoint fully-enabled and ready to go with no ability for a tenant to disable the feature on a tenant-wide or per-user basis speaks well for Microsoft’s confidence that these agents are a good thing. However, it’s also regrettable that Microsoft won’t deliver basic tenant management controls for new features. The assumption is always that customers will love the new technology and take to it like a duck to water. Unhappily, that rose-tinted view seldom endures past its first collision with the harsh realities of daily operations.
Microsoft hasn’t commented on whether licensing requirements will change. My best guess is that they’ll include Copilot agents in the Microsoft 365 Copilot license.
Lots to Test
Writing about software that’s announced but not yet available can be a mug’s game. No software is created perfectly, and we have no idea about the limitations that exist in any of the new features announced by Microsoft. Will agents reduce the problem of AI hallucinations? Is the bot-like capability of chatting with an agent in Teams worthwhile? How easy is it to extend an agent in Copilot Studio? What innovative uses will people put agents to? And will we all become experts in Python to make sense of what’s stored in Excel worksheets? Testing these and other questions will occupy time for those with Microsoft 365 Copilot licenses in the future. Whether the new features are enough to convince more of the Office 365 installed base to embrace Copilot remains to be seen.
