On the show this week – recorded just before TEC (The Experts Conference), Paul and I discuss Microsoft going all-in with Copilot Chat as it becomes more deeply integrated into Office apps; and we discuss licensing changes that enable a sizable subset of Copilot features across more licensing types, and we discuss the growing problem of “workslop”, and why many pilot programs for AI (including Copilot) don’t get off the ground.
Copilot Chat comes to Microsoft 365 apps
Microsoft announced that Copilot Chat is rolling out inside the core Office apps for all Microsoft 365 users. You’ll now see a Copilot Chat sidebar in Word, Excel, PowerPoint, Outlook, and OneNote, with grounded answers for licensed users and Bing-search (if enabled) supported answers for everyone else.
- Copilot Chat comes to the apps without a Copilot Licence (anything from Office 365 E1 or higher)
- “Premium” Copilot experience adds reasoning over all your work data in the apps
- In-App access to Research/Analysts
- Copilot licence gets “priority access” to file upload & image generation
AI Agents as Far as the Team Can See
On the podcast, we joke that we’re not sure who asked for it, but whether or not people want it, AI Agents are joining you across nearly all of the Microsoft Teams experience. Admittedly, although it’s not something either Paul or I plan to use daily (yet), these features (rolling out now) will be useful to folks – such as Project Managers working across multiple projects.
The new agents (or bots, really) are as follows:
- “Facilitator” bot: joins your Teams meetings to help out – it can generate agendas, keep track of time, take notes, and record action items.
- “Channel Agent”: an AI agent for each channel that understands the chats, files, and context of that channel. You can ask it questions about the project, topic, or have it summarize long conversations and highlight key points.
- “Community Agent”: In Viva Engage (Yammer) communities, an AI agent can auto-answer common questions by searching past posts or knowledge articles.
Thinking back, several of these sound very similar to previous AI-driven tools that were touted by Microsoft over the years, some of which didn’t last long or deliver exactly as promised. With these agents, if you set your expectations accordingly, you might be pretty impressed.
And that’s not all. There are a few more (some of which are quite interesting) features on the way:
- AI-powered workflows: schedule automated AI tasks in Teams, such as a summary of the day’s channel discussions every evening.
- AI-generated meeting recaps (in audio): After a meeting, Teams can now produce an audio recap that you can listen to, with a range of styles – clearly inspired by NotebookLM’s audio overview. We wonder if it’s powered by Microsoft’s VibeVoice model, which could explain how Microsoft can just throw-in what once would be a much-touted new feature. I’m seeing it roll out now in production customer tenants alongside my most recent meetings.
- Teams AI Library improvements: The Teams AI library gets a preview for additional languages – Python being the most interesting from an AI developer point of view, alongside capabilities for agent-to-agent and MCP communications.
Read more on Microsoft’s Tech Community
AI-Generated “Workslop” is Killing Productivity
A recent Harvard Business Review article cites research into something that you might have seen for yourself (or at least suspected). Lots of folk are using AI tools to generate content that just wastes other people’s time.
The article immediately reminded me of some of the posts on the Marketoonist site, in particular AI written, AI read from back in 2023. It’s worth a click – but the short version is – people are using Copilot to “workslopify” a few bullet points to pad out their emails. At the other end, someone most likely just summarizes it back to those bullet points.
I don’t think anyone who listens to the podcast will be overly surprised that “workslop” is a productivity killer. If, however, this shocks you and you post copy and past ChatGPT-written posts onto LinkedIn daily, then all of us kindly ask you: stop.
Read more (if you want to pay, anyway) at HBR.
And Finally: Pining over your Pilot of Copilot not getting past the Pilot and into Production?
MIT Media Lab report that there’s a reason for this. Following the previous sections, AI written, AI read, rather than pad out my summary so you can summarise it, here’s just the key info:
- Lack of retained context and improvement for generic (or even Copilot) AI tooling – Copilot is being challenged by competitors, often smaller ones.
- Little Pilot to Production Movement
- Shadow IT usage of AI is increasing (BYOAI? I didn’t get a chance to use it on the show, but feel free to use and abuse).
- The winning formula is as follows: small, targeted wins – narrow scope = win, AI painted on everything = doesn’t move past pilot.
- Emerging vendors struggle in enterprise
I have seen this trend myself in the market – financial services firms, for example, hiring AI teams focused on the really valuable AI use-cases and then more generic IT teams rolling out Copilot for a subsection of users.
Read more (Link to PDF at ArtificialIntelligence-News.com)
Join us in two weeks time for post-TEC news, and with any luck, Bastiaan Verdonk will join me and Paul to give us the best of The Experts’ Conference. Until then!
