Introduction

Yeah, so most of the time I try and figure out basically what is new that I'm trying with AI that would be worth sharing with the room. And recently I tried something new.

What’s New in My AI Workflow

How many of you know what the concept of MCP is, Model Context Protocol? More than the number of engineers, yes. That's good.

Understanding MCP (Model Context Protocol)

Giving Agents Actions via External Services

So model context protocol, or MCP, is a way, is a protocol to basically get your agent to go and connect with a whole bunch of external services, which then means that you can give the agent actions.

Building a Personal Operating System with AI

Meet Nova: My Chief-of-Staff Agent and Its Tools

For those of you that were here two weeks ago, so I have what I now call my, well, I have two names. It's my personal operating system. I also call it Nova, which is my chief of staff basically has like 230 different tools.

One of them being like I live called someone in the room after a presentation type of stuff. Cause my agent can go and place phone calls.

Um,

AI, Management, and Decision-Making

And that is kind of what's happening, by the way, on the whole manager versus employee bit and how all of that is changing.

When AI Becomes the Most Capable Mind in the Room

I have a lot of thoughts there because one of the things I would invite everyone to think about is what happens when none of the people in this room or nobody in the world is intellectually superior to the AI? What is the rational thing that you end up doing? in your relationship with an AI when you think about, okay, are you gonna do the, am I gonna do what you tell me to do because that is better for me, or am I gonna tell you to do, which is technically inferior to where I want to end up?

It's an interesting bit that comes through.

Always-On Capture: Limitless AI

But all of that to say that the new thing that I ended up with in the last few weeks, it's been about a week now, is this thing over here. Has anyone heard of Limitless AI?

Only two. It's three. There we go.

Why Continuous Recording Matters

Limitless is basically a mic. It just records everything you do all the time.

Transcripts as the Lifeblood of an Organization

And one of the things that I've come to understand over the last two months, I'd say, is that transcripts of conversations you've had And transcripts of an organization end up being extremely valuable to an AI. And when I say extremely, they are literally the lifeblood of the company.

Because most of us work remotely, or at least quite a bit of the time remotely. Most of the meetings you will have are remote in one way or another. The transcripts are there.

From Raw Conversations to Project Context

that when you have an AI agent that is able to look at what has been said, it defines the scope of the types of projects you're working on. It defines the tasks. It defines the deadlines.

It defines even what everyone in the company has expressed in terms of their own opinions on a particular thing.

Beyond Human Limits: Context Aggregation at Scale

Now as humans were fallible and we cannot or it's hard and the quality of a good project manager or manager in general ends up being like how much can you take into account the context of everyone and then make a decision based on what is the what are the actual KPIs the project. But the AI doesn't have that problem and so.

Performance Gains from Transcript-Aware AI

One of the things that I've been doing is giving my AI access to as many transcripts as possible. And it's now at the point where it simply performs dramatically better than I do.

Proposals, Memory, and Better Outputs

When I write a proposal, if I give it the six transcripts of conversations that I've had with clients, it does dramatically better than I could have because I can't keep six full conversations in my head. I will forget so much of what happened in conversation one, two, three, four. Maybe if I'm lucky, I can have most of the last conversation still in my head.

The AI can do much more.

Live Demo: Turning Meetings into Deliverables

Querying Recent Transcripts for Takeaways

So I wanna just show you a little bit of what I can do now. So with stuff like this, I can now do, can you go through the last, whatever, hour or so of conversation, the meetup that we just had, and maybe give me a breakdown, a takeaway that I can give everyone in the audience about the two presentations we just went through.

Ah, and I forgot one extremely important thing, that is that I haven't actually launched my AI assistant yet, so this is gonna be 20 seconds of extremely awkward time.

You can't see anything? Hey, okay, one sec. Why is it black?

Why, where do we have to screen people? And of course, I do the exact same thing again. I'm just going to go hide this for a second.

For those of you that are keen-eyed, I actually have a command that I call clodded. because it's an automated shortcut which launches Claude with the flag dangerously skip permissions because I just don't like it asking me for anything. I just want it to go and execute.

Okay.

Inside the Personal OS: Playbooks, Tools, and Permissions

So what I'm asking for it here, so I've got this whole personal operating system set up. It's got my playbooks. It's got access to all of these different tools.

And what it's gonna go and do now is first it reads its own permissions that actually in this case, well, I forgot to say, I think I actually told it about the, we'll see if it actually figures it out by itself.

I'm interested now because I've got the general transcripts which are from the meetings that I will have had during the day.

But I also have the limitless, okay, there we go. Limitless AI transcripts, which are the things that are coming here.

Finding and Merging Multiple Transcript Sources

And it knows that it has multiple tools to go and find transcripts. So in this case, it seems it's actually figuring it out. It's now going through the different transcripts.

Let's see if it... I can see there are recent lifeblocks including Stefano's presentation and how I will change the role of manager and now it's going through the rest. So it's found Stefano's presentation.

Hopefully it'll now find mouse. Apologies again for the voice by the way. I know it's going back and forth.

Yes. No, it sinks to the cloud and then this takes it from the cloud. Okay.

Demo presentation. It's now going through the rest of the logs because obviously it's kind of chunked this itself in its memory.

Let's see how far it gets. It provides the MCP. Okay, let's figure out where it gets to. It's a little bit more.

Automated Meetup Breakdown and Slide Plan

There we go, Meetup Breakdown.

It actually has figured out, so it's put the names on people's as well, so Mel's presentation, Building Custom AI Agents with Microsoft Copilot, the topic, the takeaway, Stefano's presentation, how AI will change the role of a manager, and some questions raised, and then the overall theme. Now can you take all of these transcripts and maybe build me a presentation that I can give to everyone that was in the audience with all the key takeaways so that they can walk away with it basically. What I really want is a slide by slide breakdown.

You have to be extremely precise I asked it to build a presentation it would go and try and build a PowerPoint presentation it's not great at building PowerPoint presentations I just want the content because I can then put the content back into gamma and then we have an actual thing we can go and work with. So it's going back through the transcripts now. Oh interesting so now it's actually looking at detailed so I wonder before. So here it was taking the content, but it was not actually looking at the full transcript.

So now because I'm asking it to really give the takeaways, it's looking at the actual things that have been said at each presentation in order to then build a slide by slide breakdown.

And it's always interesting that with live demos as much as this normally would have taken hours the fact that takes me 20 seconds or 30 seconds really feels like dead time in a live demo it's it's just like I wish they were faster. But you can see how it works and.

I had a and I'll actually show you a video right after this of the AI weekend that we were hosting last week and a half ago. So I had three and a half days of delivering an AI retreat for some executive 30 executives in Scotland three and a half days had this thing turned on and then at the end. I just asked my AI assistant to go and build me a document that would summarize everything we went through on the weekend for the 30 participants to be able to get all their takeaways in a way that would be directly relevant to them.

And about 20 minutes later, I had a full gamma breakdown. It was in this case, it was like a pretty big one. It's like 15 pages or so of stuff from three and a half days, but with very direct examples of everything we'd gone through and directly relevant applications.

So where are we now? Ah, there we go. Um, I love how it's gone.

Building the Deck in Gamma

It's figured out that Gamma was the event sponsor because I said that at the start of the presentation. What does it come up with? Designed to work in Gamma, your event sponsor. There we go.

But this is, it's the type of stuff that happens that the AI does that you don't think about. We had, and actually I'll start the creation of the presentation here. So it actually understands here

Here we go it understands because I have this temp context folder my assistant knows it's there and so I can go and just use that and it knows to store any presentations in there and here I'm going to use gamma just going to do copy paste. We did presentation here right so I'm just going to do preserve the text at this point. I actually only edited at the very end because most of the time it's just right.

Anecdotes from the Field

As it's building a presentation, which you'll see happening live, I want to share one moment we had last weekend and then I'll show you the video of last weekend at the same time.

The Phone-Call Demo That Failed—And Then Surprised Us

I went through this and at the end of the demo, what we ended up doing was I got the AI to try and call the co-host we had in the weekend. And the demo actually failed, because it somehow couldn't find the number of the person in my address book that I had hooked up to the AI. And so the AI came back and said, sorry, I can't find Jeremy's number.

And right after, and it was actually talking back to me, because I also use 11 Labs for it to speak back to me, it then came through and actually said, at the risk of stating the obvious, Jeremy seems to be standing right next to you. You might want to walk over to him and ask him this question yourself.

Lesson: Context Turns ‘Dumb’ Into Insightful

Which was a really weird moment, but a really important one to the first point that I started with here, which is when you give the AI context, holy shit, can it figure things out way more than what we think it can.

And the whole game ends up being giving it enough context to be good at what it can do. Most of the problems you will end up running into when you think the AI is being dumb for a specific thing stop for a second and ask yourself if the only information I had given to an Intellectually capable person was the information I've now given to the AI Would they have been able to answer that question without knowing anything else about my life? If the answer is no, the problem is that you didn't give the context.

It's not that the AI can't do it so here

What do we see, where do we end up?

Evaluating the Generated Presentation

We've got 20 slides now going through. Key presentation takeaway, okay. Oh no, okay, it's actually, it's not too bad.

It's just that it's done content and stuff like that. I didn't want those headers, so I should have looked at that.

But you've got tools and practice, strategic transformation, co-pilot studio, building custom AI agents, evolution of management, coordination to enablement. Part one, building custom AI agents. Enterprise AI challenge Microsoft co-pilot studio capabilities here We're to deploy custom co-pilot.

So all of this stuff was part of the the presentation as we went through It's not perfect here. And I again I would have gone back to the actual brief and I would have told it just like give me the slide by slide break I'm not don't give me the headers because that's the thing that's kind of Going wrong here and I could actually do this here Can we remove the headers because they don't make sense here?

I? or at least the indications like slide 10 or when we talk about content and stuff. We just want to focus on the actual content in the real presentation.

This is the other bit that's been a recent update on Gamma is that now you have the AI agent on the side and you can actually ask it for updates on the slide itself. You can also do a whole bunch of things.

Why Transcripts Change the Game

Okay, so that was the demo on... the power of transcripts and I would really I would advise you try out and get as many transcripts of anything that you have at any point and then basically figure out how much you can do with those transcripts.

Speed, Quality, and Scale in Knowledge Work

It transforms everything and I've gotten out to the point where entire proposals are done within five minutes that otherwise would have taken half a day or a full day because when you have the transcripts there, I can just do it much better than I can anyways.

What do we have here? Okay, so it's done a few... It's not done them all yet.

I would go through. Okay.

Q&A Highlights

Any questions before I actually go?

And I did want to show you the video of last week because it was interesting in order to get you to understand how much of a transformation is actually happening right now.

Yes.

Privacy and Compliance Considerations

Yeah, so privacy concerns here There are two answers to that question. One is there are similar devices that store locally. You can do with that what you want.

Mine stores through the cloud. Maybe if you're in a big corporate context, you can't use something like this. They'll say, no way in hell. I get that.

Totally fine. I'm lucky. I'm the CEO of the company. I can say, yes, we can use this stuff. But

Most of those companies, at this point, there are still a few that are holding out. Most companies are happy with transcripts on Teams calls, Zoom calls, or things like that. You get a lot out of that.

The real challenge is going to be, how do you make sure that your AI agent has access to everything that's there? Because it's sitting somewhere on a Microsoft server, and you need to figure out how you can access them. And that is where everything else then falls downstream. Yep?

Have you got the overall code and the whole AI thing? Sorry, documented. No, not at the moment. It's something that I will be doing more of.

Yes. Yeah, yes. Yeah. Yeah.

Access Control with MCP and API Keys

So I used a lot of MCPs. How do I know that other people in the firm won't be using or will be accessing stuff they shouldn't be able to access?

1All of this relies on API keys. The API keys are authenticated to a particular user account. That means that it only has access to whatever that user account has access to.

Sorry? I own the data. It's 100%.

Now, I'm going to be here afterwards as well, and I don't want to wait too long for the pizza, because I know the pizza is actually already there.

AI Weekend: A Glimpse into Rapid Transformation

So I just wanted to finish, actually, on this video, if it comes along. This is the first... It'll be half promotional. It's four and a half minutes, but I really think it's worth sharing, so hopefully that'll be it. You tell me if it was adding to the event or not, please, at the end.

Rarely do this, but I thought it was worth sharing. So this was... This was the AI weekend that we hosted a week and a half ago in Scotland with 30 executives over there. So let's see. External, there we go.

There's nothing to talk about, James. We're down from two out of our scenes. Riding on a wave through the pines. There's no one to help us. Yes, it's AI music.

Participant Reflections

I came here this weekend to figure out how to optimize the way that I'm using AI in my practice to open up new areas of business development and ultimately to help lead our firm to be a leader in AI adoption. Each individual hour has been worth the price of admission but the most valuable bit of this so far has been to work interactively with the trainers and the course instructors Because their depth of knowledge of how to operate and how they can connect with each other is beyond any knowledge that I have or any knowledge that I'm aware of.

By Coffee Breakers the first day, I'd already realised that the value of the course was achieved. I know there's a lot more room to go, but I've got increased motivation to now really devote my time to understand how I can best use AI and tools in the business benefit and my family. And that's likely to be the best investment that I've probably made in the last 10 or 15 years.

You can still look back at your life and maybe that day, for a certain time, you made a good choice and invested well, whether that's the time or not, it's something else. And I think this is one of those times.

I am using AI in daily basis, but I didn't have the time to dive deeper and create something more fascinating. I mean, I was astonished when

Joshua, at the end of his digital garden, and he did the live call with Dimitris. And then, when he lied down with that, I was like, OK, this is what I'm going to do for myself. I really wanted to dive deeper. And I was very curious to see the next step.

and to get more ideas on how current organizations in the scale of countless people really massively adopt and embrace practical reality. transform the way that you work. You know, you're going to finish work on Friday and you work in one way and you're going to go back in on Monday morning and the way you work is going to be completely transformed for the better and you're going to be able to do so much more, in so much more depth, to so much better quality, so much more quickly than you ever could have before.

And that's going to happen literally over a weekend. Like, I never really believed it before, but now I know, wow, I can't wait to get into work tomorrow to start doing it all. Now let's go build something.

I've done a lot of executive education. I've done the CEO, the courses at ESO, I've done the MP at ESO, I've done a couple of courses at Stanford. They were all great, but none of them lift the roof off my head like this did. Absolutely amazing. You're going to have your mind blown.

Now, I know how promotional that was. I am proud of the weekend, to be very clear.

Closing Thoughts

Bridging the Gap Between Skepticism and Practice

But the thing that I keep coming back to here is how massive the gap is between when you understand how to actually use this technology and when you don't. And a lot of the disconnect that is currently happening when you have people saying, oh, it's hype, it can't really do these things, we're getting too excited, actually comes from the fact that people don't understand how to use the tech.

And I know I'm saying this and some really credible people are talking about how it can't do stuff, but when they're actually shown, and some of the people were on the weekend, When they actually see what is possible to do and they start to put the pieces together, it's just we are going to look at a few years of extreme weird stuff coming out. And this is just a starting point.

Thanks and Next Steps

Thank you very much. Thanks.

Have some pizza. Have some beer. Come and talk to us.