Microsoft Ignite 2024 - circling the AI drain on enterprise productivity

The dizzying rise of Microsoft Teams has made the vendor a leading player in the digital teamwork market, so I was looking forward to hearing how it plans to advance the state-of-the-art at its annual Ignite conference, which opened yesterday in Chicago. Yet despite a mammoth catalog of over 90 new products and features in the 'Book of News' that Microsoft traditionally issues for this mega-event, I was disappointed to see digital teamwork effectively sidelined in the vendor's headlong rush to embrace AI. So instead of being a product news report, this article has now become an op-ed, about Microsoft's focus on AI as a technology, rather than as a business enabler.

Don't get me wrong — I'm not denying the massive significance of AI. It's going to have as big an impact on the world as the advent of the Internet and cloud computing combined. But as with those earlier technologies, the rapid change they bring is going to come along with a lot of false starts. We are currently in that phase of the technology when you get the most false starts, when everyone makes hasty investments for fear of missing out. Large incumbent vendors like Microsoft are particularly prone to these errors of judgement — as is borne out by its history with both the Internet and cloud computing.

I say all this even though there were game-changing announcements yesterday for many users of Teams. Announcement 6.1.1 in the Book of News heralds the imminent arrival of meeting transcription for multilingual meetings, with machine transcripts available to meeting participants in any of 51 spoken languages and machine translation for any of 31 languages. The translation capability will also be available next year for live captions and transcripts, and for automated intelligent meeting recaps delivered to participants after the meeting in their chosen language. Clearly this will be a massive boon for those who have to take part in meetings held in other than their first language, particularly when they have poor or no skills in the language spoken. Even more thrilling, listed as part of announcement 1.1.1 in the Book of News, is a new Copilot Interpreter agent for Teams, available in preview next year, said to be capable of simulating each participant's own voice while interpreting what they're saying in any of nine languages. This is the stuff of science fiction — AI that turns you into a fake polyglot.

But here's the issue I have with an announcement like this. Probably the best way to make multilingual meetings productive for the people involved is to do as much communication as possible outside of the meeting — send briefings out beforehand, have a clear agenda and goals, maybe not bother having the meeting at all. Using AI to ensure that even more people can spend their time sitting in poorly planned Teams calls isn't going to improve anyone's productivity. Far better to have an AI agent responsible for making sure that all the necessary briefings were distributed in advance, that people had read them and submitted any questions, that there was a clear agenda and purpose, and that only those who needed to be there were invited. If you want one of those, you'll have to build it yourself, although Microsoft has at least made a start, with a Copilot in Outlook that helps schedule meetings and draft agendas, while a Facilitator agent in Teams, now in preview, takes real-time notes and summaries of video meetings and chats.

Grounding AI

My overall impression from reading through all 90+ of yesterday's announcements is that someone said, 'Prioritize AI'. In fact, we know that's what happened. Satya Nadella, CEO of Microsoft, stepped out on stage yesterday in Chicago for his keynote and stated:

Today, I want to focus on AI and this transformational power as it drives growth in business. It improves efficiency; it improves operating leverage. And to do that, we are building out three platforms. Copilot, Copilot Devices, and Copilot and AI Stack. That’s it. Those are the three platforms.

In fairness, he did also preface that statement with the remark that, "It’s not about tech for tech’s sake, but it’s about translating it into real outcomes." But we also know from the investigation published last week by Business Insider on what it called Microsoft's struggles with Copilot that the focus on AI has led to investment being pulled from other product areas, including Teams, according to an internal memo shared with the BI investigators. Was real-time multilingual transcription and interpreting really the next top priority for Teams development? Or was it chosen as the most AI-intensive feature that would best align the product with the corporate mission of building out Nadella's three platforms?

I was surprised that yesterday we didn't hear something we frequently hear from other digital teamwork vendors, who invoke their proprietary graph mapping of team members, tasks and goals, plus their relationships to each other, as a crucial tool when working with generative AI to contextualize instructions and responses. I find this particularly odd, because Microsoft was the first vendor to start talking about such a graph database, more than ten years ago now, when it was called the Office Graph. This is exactly the kind of thing that would give Microsoft an edge when it comes to grounding the prompts that users give to Copilot in order to find information or initiate an action. But the Microsoft Graph seems to be entirely absent from the action. Perhaps it was considered too complex to deliver. Instead, AI is being let loose to surface information or automate processes, but with users themselves — or their administrators and developers — having to custom build the prompts and guardrails that will ensure the AI comes back with trustworthy answers.

This lack of grounding for its AI is at the heart of the criticism levied at Microsoft by Marc Benioff, CEO of its competitor Salesforce. Writing about the rise of agentic AI and what it means for enterprises a few weeks back, I said I expected to hear Microsoft's side of the story at Ignite. Yesterday did see Copilot advance into agentic AI, with the launch of several ready-made, purpose-built agents to do tasks such as find information in SharePoint, respond to self-service requests for HR and IT tasks, and oversee projects, as well as the Teams agents mentioned above. Customers can also build their own autonomous agents using Copilot Studio, while various developer tools were introduced for building AI apps and agents in Azure.

Salesforce vs Microsoft

But the core of Benioff's criticism still stands - that Microsoft provides a less robust framework for agents to make sense of information or carry out actions. While Benioff certainly has a point, I think the truth of the matter comes down to the type of data the two vendors work with. As an enterprise application vendor focusing on CRM, Salesforce has a highly structured metadata model that it is able to use to instruct and direct its AI capabilities. When it brings unstructured data into that model, it brings it into that very structured framework.

In contrast, Microsoft is largely an enterprise productivity vendor. The bulk of its data has always been unstructured, and even if it's increasingly moving to the cloud, most of this remains on-premise. Even with the Microsoft Graph — to the extent that it can be applied — the vendor is dependent on its customers doing the work to impose structure. As the Business Insider team found, many customers have not even been applying user access permissions with any rigor, resulting in Copilot frequently surfacing data to users that they shouldn't have access to. Retrospectively applying those permissions requires a lot of hard work, as does the addition of useful metadata. Maybe this is work that can one day largely be automated by agents, but until that preparatory work is complete, it's difficult for AI to derive useful value from unstructured data. By the way, this applies equally to any of the same data that Salesforce might want to bring into its AI framework from, for example, a Sharepoint data store — it doesn't magically acquire structure just by virtue of bringing it into the Salesforce framework. The customer still has to do a lot of work first to get it ready.

Also unlike Salesforce, Microsoft is spread far more widely as a vendor. It's a hyperscaler cloud platform as well as a productivity and teamwork applications vendor, as well as an enterprise applications vendor, as well as a database and infrastructure vendor, as well as an operating system and hardware vendor — it has even brought back network computers again, now called Windows 365 Link. It's trying to leverage AI across every one of those bases, and this technology is massively expensive, especially at the infrastructure layer, requiring massive upfront investment before realizing any revenue, in the same way that cloud computing did in the previous decade. The company has therefore made a big bet on being able to monetize AI by charging a hefty user license fee for Copilot, in response to which customers are demanding measurable returns on their investment.

Picking winners

All of this activity is fueled by the recognition that everything is moving insanely fast. Nadella started his keynote by recalling the launch of Windows 3.1, more than three decades ago in the very same conference center, which secured Microsoft's dominance of the PC industry and triggered the long-term decline of IBM. At the time, Moore's Law ensured that microprocessor performance doubled every 18 months. Now, says Nadella, "With AI, we’ve now started to see that doubling [in performance] every six months or so."

People tend to underestimate the compounding effect of scaling at that pace. Fail to act and your competitor will have a 10x advantage in less than two years, a 50x advantage in less than three, and a 100x advantage six months later. But if you make the wrong choice, you'll not only fall behind just as much, but lose your investment too. Deciding what technology to adopt and when to adopt it becomes critical.

Microsoft clearly believes that by putting as much AI technology out there as possible, it is helping its enterprise customers pick the winners that they believe will help them achieve more. But as I mentioned earlier, productivity isn't just a matter of doing work better, or faster, or more efficiently. It's also a matter of what work you choose to do. The danger in the rush to embrace AI is that the investment will be wasted on automating work that doesn't actually produce value. Yes, there are short-term wins to be made — Nadella cited several customer examples in his keynote — but there are plenty of opportunities to add new bottlenecks and glitches too. My sense is that Microsoft has too much faith in the technology per se and is not thinking enough about how to help its customers make the most productive use of it.