MVP Summit - Quickfire Questions with one half of Copilot Connection!

speaker-1: So welcome to a quick fire round for securing the realm this week. ⁓ Got a special guest with us usual ⁓ so Kevin, want to just kind of introduce yourself, tell us a about you and we'll kind of get straight into the questions. speaker-0: And it's very funny being this side. So Kevin McDonald.

I'm the co host of the Copilot Connection podcast, ⁓ Microsoft and Copilot and Copilot extensibility. And I'm lucky I get to work with Josh Avanade as well. ⁓ speaker-1: Absolutely. Pleasure, of course.

⁓ No, it's really interesting to have this conversation because ⁓ from my side, looking at Foundry and then you've bringing the Copal extensibility. So you've got both sides of the coin really from a Microsoft AI standpoint. So let's start big. Where do you see AI going?

What's your hot take for the year? speaker-0: I'm fascinated where the kind of skill side, a lot of the things that's happening with Claude, co-work and things, is opening up AI to more and more people. rather, you know, love the MCP, I love, you know, you've got A2A, all these lovely protocols, but the average person is going to go, yeah, I have no idea about this. But I think that ability to write skills is a very natural thing for people.

And I think this evolution that we're seeing of people being able to make more of that is going to start flexing more out of the developers, more out the dev world, and obviously coming from the copilot studio world, it's kind of opening up to lots of people. But I think this will see that transition. And I think that's going to be the fascinating part to it as well. So there's speaker-1: skills gap and it's being addressed with skills in the literal sense.

speaker-0: And I think it's just going to be fascinating to see where people get with it. And we do a co-pilot far side chats every month. We had a poor Yoho from GitHub and the thing she was talking about, you know, I was expecting it to get quite technical, but she wasn't talking about all the things like generating spreadsheets and work on that, that she was doing with her coding agents with skills ⁓ running in VS code. I was looking at going with a little bit of an interface change, anyone could be doing this.

This is that kind of world just with a bit of thinking. I think so we're going back to the original question, where do I think AI is going to go? I think we're going to see it open up to more people of thinking of things that they can do that's a bit different. So I'm really fascinated with that.

And I think the other slight world, and I don't quite know how, ⁓ exactly. But, ⁓ Jamie Teevan, who's the chief scientist at Microsoft has been talking a lot about the we of AI. And I think right now we, it's kind of a personal thing. You go and chat to an agent, use co-pilot, you write some code or it's built into an app, but more and more things like facilitator that that's there in teams that sits there within the meeting, kind of nudging you onto the right thing, listening to what you're doing, capturing that.

we're going to see more things where it happens, channel agents in teams, things that sit there while you're doing stuff. I think, yeah, I look at the Alexa I have, sorry if that's triggered anyone's around there. I look at that, we have one in the kitchen and a number of times when we're kind of eating with the family and someone's got a question, we'll ask that. Now, I'm going to terrify people slightly, but imagine that could join a conversation.

We could have an agent running during this conversation, listening to us. trying to suggest things, open things up. There's going to be a way that that's not intrusive. It knows when to join in.

You don't have to actively trigger it. But I think we will see that evolution as well. It'll really interesting. speaker-1: So kind of an ubiquitousness to it all, if that's the right word.

Ambient computing. speaker-0: Yeah, I think we'll be fine. In fact, if I think of your Dungeons and Dragons one, I actually saw the photo from your first securing the realm at Scottish summit to be able to the thought of having like a dungeon master that sits there with you. That was kind of the first steps we're seeing in that we bringing into the conversation, I think we'll see more and more things.

And I think we'll also see things we haven't thought of before. And that's the bit that some more go. we could use AI for this or we could bring that into the conversation. I always love those moments on there.

I think the technology is there. It's our habits and actually thinking of it that will be the big thing. speaker-1: Yeah, I think I was watching a podcast, Hanselman, it's Scott Hanselman, of course, where he interviewed Peter Steinberger, founder and creator of OpenClaw. Not something I personally use, but one of the things that kind of he kind of said about it was he just put it together.

people hadn't put it together yet. And so, like you said, the technology is there, but it's what we then figure out to do with it next. Yeah, this point. speaker-0: Absolutely.

it was funny when OpenClaw first came out, Gary Trinder was telling me about it. But it was actually the way they were kind of structuring, I can't remember what the different parts of it called, but basically the skills within the main agent itself. He was showing me how to structure that. And I was like, wow, this is just a way of thinking.

I have like a vision one. There was another name for it, I completely forgotten, but the way it kind of brought the different things through. really made sense from there. the kind of back into it was as fascinating as the ability to use it.

And I'm saying to you, haven't set out. Why? Because I'm terrified of what it can do. I don't feel I know enough about it to control it on there.

So at some point, I will might kind of run it in some isolated way with this little, I'm thinking about sort of all this, I know everyone's been buying the Mac minis, been thinking about getting something that's a brand new one. doesn't have anything turned on. Yeah. It's just used for that.

So we kind of have a little bit more control over it. But I think the guard rails for that are going to be fascinating. speaker-1: And moving on to that actually, like, how do we secure this stuff? How do we govern it?

speaker-0: Yeah. And I think there's two different questions there. Yes. intentionally that they're put together, but there's two different stories.

There's one protecting it of making sure absolutely, I was going to say the barriers. know that security is always talking about not having the firewalls, not just protecting the outside, but throughout, but it's that absolute, must not do this. But then there's also the governance. What should it do?

What shouldn't it be doing? How do you kind of keep it on track? And how, how do you kind of look at what it's what is right and what is wrong? And that's a horrible way of saying it, but the kind of ethics of these things, at what point should you allowed to like, for example, do I want my open core to be listening into my dinner conversation and being helpful there?

At what point do you say, no, it shouldn't be listening to that? because we could be discussing sensitive things that we don't want to go out there and we don't, we don't want to be influenced by that. So I think the governance within there is less of strict rules, but more of how do you decide when to use it? How do you stop it from stepping in in those situations?

It's, it's going to be interesting. And I, you know, my wife is, doing AI myself a lot, she's an absolute hater of AI. She just doesn't trust any of the things on that. kind get her, but they get exactly where she's going.

And it grounds me a little bit and kind of, I have in my head, what would Celia do? What would her opinion if I was telling people about this? And I know that open clawed absolutely terrified. If I let her know, she wouldn't let me turn it on at all.

So, it's kind of getting that acceptance factor working out. I think we all need to challenge and say, is this actually a good idea? And I kind of feel we've at times gone too far and stopped. stop checking on is this good idea enough.

speaker-1: You've brought me on to an interesting question actually around, you know, looking at this governance aspect. Have you tried using something like Foundry Evaluations for that? Kind of not necessarily for open call-opix, you're not using it, but yeah, know, the use of the tool. ⁓ speaker-0: A little bit.

Yes, certainly. ⁓ I'm wearing my Jen to get in top we did a workshop at ESPC in Dublin with Chris Huntingford and Zoe Wilson. And that was one the things we were talking about red teaming within that how do you how do you try and get agents to do things they shouldn't do but also you know, what is the barriers you want for that and then using evaluations to run that on a continuous basis, look at it from different models, how does it perform from that? So yes, to a degree, I can always see feel my brain ticking over.

Yeah, this is interesting that I've kind of always done some of those evaluations to define the borders. So I want to make sure it doesn't do this. How far does it get within that? That's interesting to see how much it does do, know, what sort of things come back that come from there as well.

That would be interesting to look at. speaker-1: So kind of like a CI, CD and CE for continuous evaluations. speaker-0: I'm ⁓ speaker-1: We have a tongue twister as well at speed. It's been a real pleasure, Kevin.

Thank you so much for coming on. Cheers. speaker-0: Thank you very much.