Oakland | B Corp’s Post

Transcript

OK, AI folks, we have to talk about Copilot and SharePoint. It is Day 2 of Microsoft Ignite and I have just got off of. I talk about Copilot in SharePoint. And, umm. Yeah, we gotta get into it because like it or not, SharePoint is where business actually happens. And increasingly in the AI development space, we are going to be asked the question. In fact, I'm already getting asked the question by clients. Why don't we do this with Copilot? Because like it or not, SharePoint for many enterprises is actually where business occurs. This guy who was more excited about SharePoint than I've ever seen any human alive. But hey. You do you, Jeff Taper. That's awesome. This guy was talking that they see an ingestion of 2 billion files into SharePoint every single day. 2 billion flows get executed in Power Automate every single week and 2,000,000 SharePoint sites get made every day. So we need to wrap our heads around how SharePoint is working, what Copilot is going to be doing in SharePoint and where the limitations are and when the stuff I talked about. Yesterday with Azure AI Foundry and Auto GPT and the cool sexy agent **** that we'd love to do, where does that take over? I now know way more about SharePoint than like I particularly want or need to know. But it is incredibly important because like I say, it's where business actually happens. And for the vast majority of people and the vast majority of use cases, I suspect Copilot in SharePoint is going to be their principal interaction with. AI and it is here from December the 1st and the principal thing that they are pushing with Copilot and ohh boy are Microsoft pushing Copilot for SharePoint. If you have 50 Copilot licenses in your organization from December the 1st for a quote limited time, you're just going to be able to use Copilot Unlimited. There's just if you have 50 licenses, you can use it as much as you want. If you don't have 50 licenses, you're going to be able to. Use it 10,000 times in your organization per month before you have to start paying for more. And there is a consumption layer of which begs the question. What is actually happening when you summon an agent, what they're calling an agent? I think we need to make a separate video at some point about all this nomenclature because. Do you Word Agent is used in at least three different contexts in the Microsoft Ignite? Difference and copilot can mean three different things depending on what you're talking about. Those copilot, those copilot for sales, and then there's a different copilot that I've temporarily forgotten. So we just. Focus. Copilot and SharePoint has these things called agents and under the hood I think they're secretly GPT. Under the hood they are secretly leveraging Open AI's function calling API. To augment. Normal interactions with GPT 4 O with Http://requests to the content that is in your SharePoint sites. And I think this for two reasons. One, because weirdly, I've spent basically all of today getting way too into the SharePoint API and how that works for different, more boring reasons. For actual work that I was doing and how that interaction happens and how everything is actually an HTTP request under the hood. And then the. Other reason that I think it is because when in the talk I was just in, they showed the dot agent file, which is the big piece of JSON that characterizes your agent. So when you've made one in SharePoint, when you've made yourself a copilot agent in SharePoint, you get a dot agent file, which is this JSON. You can pick that up. You can give it to somebody else and you can then they can use the agent that you've made. When they showed that I was able to in sort of five seconds that they went, this is some JSON. Look at it. Pick out a whole bunch of the guilds and the access and the addresses that I've been mucking about with all day to make those requests. And so that's what I think they're doing is they're actually augmenting a GPT 4 O regular API request with that HTTP request, giving it all the permissions and the service principles and the Guids that it needs to actually make that request to SharePoint. So what does this mean for us? What does this mean for AI developers? Well, principally it means that you, you just need to know how it works. And you need to know where it stops working because what we're getting here is RAG. We're getting retrieval augmented generation. I think that it's not a vector store that we're getting. What we're getting is a combination of, if your document is small enough, just taking the damn thing and piping it into a GPT API request. And then we're also getting some of the proprietary graph sentiment data search that Microsoft have developed. For the real time graph rag that can occur over files. That has limitations. OK, that that has limitations. It runs out of beans at certain file sizes, and it also runs out of beans in certain file types. For instance, it absolutely cannot read a SharePoint list. So if you have something in a SharePoint list, these guys can't get to it. So you're going to have to do something with that, and you're just going to have to be ready for all of the things that normal users are going to be able to do. Normal users are going to be able to just make an agent on the fly. And so if you're not careful and you don't administer your SharePoint site well enough, you're just going to have like a squillion different agents, all of whom are good at looking into one folder. And so they're going to have to be very carefully managed. People are also going to be able to augment these with power, automate flows. So where in auto GPT land or like say Lang chain land, lane graph land or something like control flow, we would give agents tools and then allow them to. Execute code. These guys are not going to do that. They're going to be given the ability to execute power, automate flows, which is slightly different and in my opinion slightly jank in its own way. But you get good integration into reasonably routine business tasks at that point. And they're also going to integrate into Teams really well without you having to mess about with the Teams API, which is a horror in and of itself. But if you have to do something complicated, you are still going to be using Azure AI Agent. Foundry, These copilots are not yet in a place where you can have really big branching problems. So if you have a multi agent workflow, you should probably use as your agent AI Foundry or however those words go together because agents in copilot can't call other agents. So if it runs out of steam, it can't show up and be like oh I've run out of steam in this folder, I need to call my mate over here who knows things in that folder and get stuff out of there. You can't do that. That functionality is not here yet. And so we're going to have to be able to develop stuff in either AI foundry or our own implementations like control flow or to GPT or whatever it is that we do. But we're also going to be ready for the amount of gulf that's going to be produced. We're going to be able to make page creation with Copilot was a really big deal. Again, my notes, I'm going to type these up. If you are still watching the video, I'm gonna type these up, put them in GitHub, and then everybody will just be able to access them. We're going to get auto filling of metadata. We're going to get structured output, which is quite cool. So you're going to be able to have. Normal business reporting, and if you have a standard Business Report that you run, you're going to be able to just fling emails and Word documents and PDFs and all kinds of gubbins at your agent and go. Can you just read all of those and turn it into the structured report that my CEO is actually expecting, please? All of which is to say, I'm actually looking forward to tomorrow. I've now seen the word SharePoint so much that it's beginning to lose all meaning. Not that it had a lot of meaning to me before anyway. And copilot, honestly, I'm beginning to lose my grip on reality. There are so many different copilots and different configurations that I'm, it's just chaos out here and we're gonna have, there's gonna be a whole. Cottage industry, maybe not even a cottage industry. I don't know what a big version of a cottage industry as a whole. City industry. It's gonna be a freaking metropolitan industry out here of people who assemble this mess of copilots and turn it into something that's your business can actually make sense of. And Microsoft at least have some idea that that is going to happen. There is very slick integration with Microsoft per view. If you are familiar with the data governance tool per view that is going to be governance. First, many of these agents and copilots and things, but either way. It's gonna be a little chaotic. Like I said, I have run out of beans for this evening. The the, the, the Ignite will continue, but I'm going to watch it on catch up. There's some stuff on connectors in Copilot Studio which I wanted to get to this evening, but I'm just not going to have time to. So I'm gonna watch that and catch up tomorrow and we'll see how that goes. But tomorrow is way more to do with their multi agent calling and what you're going to be able to do with Azure AI Agent Foundry. So I'm much more excited for that. Follow me if you want the updates.