Introduction

Over the last few months, actually at this point probably two years, I've been building a productivity system. And I mean, for anyone that knows me knows I really like kind of tweaking my systems.

I like AI, weirdly, hosting this.

And so I try to use AI in different ways to actually be productive specifically for non-technical people.

Integrating AI into Productivity

And I've developed this thesis a little bit where as the AIs get better, The limiting factor for an AI to be useful is going to be how much context it has on you, on how you like things done, in order for it to be executed in the right way.

And so for the last few months I've been building up a whole bunch of different what I call playbooks and context on how I like things run. So I'm actually, I'm using Obsidian here. I've created what I named my chief of staff.

I've named them Nova. And I've given them a whole bunch of different context.

I've told them you've got access to playbooks, guides, context on me, templates on how I like to write follow-up emails, how I like a proposal written, how I want research to be done.

All of that is kind of defined, and you can see them here.

So if I look at examples, you can see, so I've got call transcript examples, I've got emails, and how the follow-up emails have been written. I've got investor updates. How do I do investor updates as examples?

All of that has been categorized and I've now fed all of that into the AI.

Using Claude Code for Operations

I use Claude code, which I know they did a terrible job at calling it Claude code and they did another terrible job, which is hosting it in a terminal. So I'm sorry if that feels scary because I know that non technical people will not be familiar with the terminal in any way, but hopefully you'll see how easy this is as I go through.

So I created another command, by the way, which I called clodded instead of clod, which triggers clod with a permission, which I don't know anyone that's used it, which is a mode which is called dangerously skip permissions, which is the thing you're not supposed to be doing.

It means it basically just executes whatever it wants to execute without asking you for it. So it's going to be able to access my machine.

And so the first thing I'm going to do just to show you what run books would do here. First thing I'm going to do, let's run through my, um, learning log for the day.

So I'm using speech to text as you can imagine. And so it's just pasting that in the terminal.

There's nothing much, right? But what is happening in the background?

Running a Learning Log

is it has been configured to look at my chief of staff, which means that it knows to go and look for my playbooks. So it's first looking at all the playbooks.

What does it say?

Okay, here we go.

So ask Joshua one question at a time.

So I'll show you here on the screen my learning coach.

My learning coach is basically a process I use every day. It's the way that I go through my end of day and it asks me, it takes on the persona of an executive coach.

It goes through one question at a time. 1It makes me rate my day on four KPIs, which are pride, growth, wastefulness, and fun. And it's the way that I keep track of my day to day and I chart that and I have fun with all of that.

It makes me figure out if I have a bad month or a good month. And then it tries to look at previous learning logs and surface something interesting from them based on a trend that it detects.

So the first thing.

Okay, so it's actually found the learning log. It's now created a list. It's going through.

Let's start with your pride rating for the day.

Okay, on a scale from minus two to plus two, how would you rate your sense of pride on a day? I'm going to say 0 at the moment because it really depends on how much of this demo is going to pull off by the end of it.

And just to accelerate through it, I'm going to give you the rest of the ratings as well. So my growth, today I'm going to say plus 2 because whatever happens, I will learn from it.

My wastefulness. decent today. I had to and so I don't know if it is in the crowd at the moment, but he borrowed me his account to be able to go and use a bigger model than otherwise would be used in order to pull off this demo. So I that took a little bit time setting up now the last one the fun part.

I think there's going to be fun either way, even if it fails is going to fail spectacularly. So we'll see.

And I just realized I didn't actually give it ratings, so that means pride zero, growth plus two, wastefulness zero, fun one.

Okay.

So it's now, so I filled in, the life KPIs. Now it's going to go through the next step.

It should ask me for any important decisions made today. Let's see. Oh yeah, well it's doing that.

So no, what important decision did you make today? Nothing much other than the fact that I'm doing a live demo on somebody else's account and I hope I didn't mess that up.

Last bit, what am I grateful for today? The ability to be able to do these types of demos in front of people that really kicks my ass to try and make sure that I get my act together.

The Learning Log Process

Okay, so after this, we're gonna go on to the next bit. So that was my learning log playbook, right? So it goes through my learning log.

This is how it would go through the end of the day. It's going to go and save this in Notion now. which I realize I should have prepared because I don't want to pull up my notion with everything in it at the moment.

But you can see it is saying search notion for connected sources, learning log. And this is the thing that the models are not perfect at yet, which is that it does take, I've connected all of these systems to my cloud code, but it's still taking like seconds, minutes in some cases to go through.

So here it's malformed a search. It's going through a few errors. Let's see if it figures it out.

1So the interesting thing here is in the learning code, I have actually given it the learning log database idea because it does have this mistake from time to time. So here, this is a good example of the AI not being perfect because it should know exactly which database to go and store this in.

Okay, well it figured it out in the end. Provide the most valuable question or observation based on all the information that it has. So it's now gotten through.

Okay, let's see. Your most valuable insight of the day, the fascinating contrast between last week's triumph and plus two across all KPIs.

I had a good week last week. Your pride rating of zero shows your understanding that real pride comes from execution, not just preparation.

There we go.

So it's gone through all of my learning logs and then surfaced that as a trend gone through. Actually, it surfaced the data even though it had failed at saving the data to begin with.

Now I want to go on to the next step.

Now it's trying to save it again. Again, live demo gods, this stuff is not perfect, but the next stage is going to be fun.

Okay, so now I think it just gave up. It's creating it in the main database instead of a sub-database somewhere. And that seems to work.

I'm gonna get already to the next one.

Deep Dive with the Executive Coach

Now let's use your editor playbook and simulate a conversation with the executive coach to go one level deeper. Let's look at my learning logs and maybe try and surface some deeper insights.

Use my demo playbook as well to anonymize things as much as possible. So it's going through another few playbooks. I've got a document editor playbook.

Um, I've got demo mode, which is as found, and now it's going to create a working document for a session with my executive coach. So let's figure out what it's going to come up with first.

Okay, so first thing in my temp context, you can see there's a document being created now.

Learning log analysis, recent entry, previous week's entry. Question that comes through from the executive coach.

For some reason it's, because I used the demo playbook, I asked it very specifically to anonymize everything, so it's named the coach Dr. Sarah Martinez. which obviously is not the name that I used in any way.

You rated growth plus two in both entries, but under completely different circumstances, one during mastery and success, the other during uncertainty and potential failures. Most executives I work with show inverse patterns here.

I am irrational.

Tell me, when you're, When you're in that space of unknown outcomes, like today's demo situation, what specifically are you learning that feels so valuable it merits plus two?

What's the internal experience there? This is a very interesting question.

So there's something about the urgency and the fact that you have to go and do this type of demo in front of a whole bunch of people that makes me try and prepare for it in a different way. And suddenly I get to do the things that otherwise I might have waited another month for.

Analyzing Behavior and Motivation

Now for the rest of the conversation, let's use the speech playbook at the same time. Oh no.

So first thing you will have seen, the AI is now annotating the document. It's not modifying it, it's annotating it.

So you're describing urgency as a forcing function that compresses time. Let me dig deeper here.

When you say things I might have waited another month for, Are these typically tasks you know you should do but lack the motivation?

Or are they genuinely lower priority items that suddenly become essential because of the demo context? That's a very good question. I would say the demo context basically forces me to get my act together in time in order to not totally fail whenever I have to go and do it.

So two things happened. I've got my editor that is editing the document underneath here. Oh, look, second bit here.

Fear-based excellence. Demo context forces getting act together. To avoid failure, suggest underlying perfectionism masked as growth mindset.

Oh shit, there's something going on here. Now that's the real insight.

You're describing failure avoidance, not growth seeking. This creates a fascinating contradiction with your growth plus two ratings.

Let me ask you this. When you're in those high pressure demo moments, what's driving you more? The excitement of what you might learn and achieve or the terror of public failure?

I think for a demo, this might be getting a bit too personal.

Let's finish there.

Real-Time Feedback Loop

But I'd love to get Adam McCann's insight on this. Can you please give him a call and ask him instead? Give him a little bit of context on the demo that we're doing.

I'm just correcting the spelling because it got Adam was it Adam McEton from what I was saying. So what it's doing now it's hooking up to my address book. It's figuring out the, um, the contact is of Adam.

Adam is in the audience. Um, it is creating a call in a few seconds, a few seconds. It might take, it might take another 20 seconds.

It's conduct condensing everything that we just have. Okay. The call, you will answer.

You can hang out. You can hang out now.

So we're going to see here the call should wrap up soon. It's actually telling me this because it's observing the call.

Oh wait, is it still? Huh interesting thing that I mean there had to be one thing that went wrong in the live demo it stopped observing the call so I had to tell it now is the call still ongoing.

Now it's going to do a perfect call just ended Adam gave some great insights here's what shared. Let's figure out what comes back. Okay so.

Can you actually vocalize this so everyone can hear. Wow, 10 seconds in a live demo is a lot.

So I just got off the phone with Adam McCann, and he had some fascinating insights about Joshua's learning pattern.

When I explained the fear-based excellence concept that came up in the coaching session, Adam immediately said, have you been on LinkedIn? Absolutely. He recognized this pattern as incredibly common among successful entrepreneurs.

His key insight was that whatever works is generally the rationalization, meaning successful people often convince themselves their fear-based motivation is actually growth-seeking because it delivers results. But he also said that deeper introspection like Joshua is doing through his learning logs is worthwhile. The timing was perfect too.

Adam was literally sitting in a church listening to Joshua's demo when I called him.

So we got a real-time validation from another entrepreneur that this tension between fear-driven performance and genuine growth mindset is something many successful people struggle with, even if they don't always admit it.

Conclusion of the Demo

So that's the end of the demo. And whilst I'll take some questions, I'll actually add one thing that I didn't think about.

Can you create an image about this entire interaction that somehow captures the essence of it? And I'll go explain a little bit about everything that happened.

So I want you to think about what all of this means. So this only works with the recent release of Anthropic, which is the one million token context model of Sonnet 4. And when I say only it kind of works I would have done.

I would have done a demo that would have been 90% as good. With the other model because the problem is that at some point it start con it has too much context there's too much stuff going on it goes through a compact. compaction stage and then it loses all the essence and it's like I was writing about this this weekend it's like you've got this thing it's really getting somewhere but it feels like it's got two minutes of memory and you're like every two minutes you're starting back up but this was the whole this really changed everything from it and so.

I have 236 different tools attached to my agent. It can do so much more than what I just showed, but you can see what's happening here. You've got an intelligence.

It can help me reason, so this is why I love this annotating approach on a document rather than anything else. I don't want it to modify the document. I want it to tell me what it thinks needs modifying so that I can then go in and here say, well, this I am going to reject. But this, I'm going to accept. Just like you would with someone annotating your own document.

This is not just creating the document. If you think about structuring a document is structuring your thoughts and trying to make up your mind on a particular point, pulling in a whole bunch of data, it can interact with the physical world.

As you saw, we placed a call to Adam, who graciously answered. I'm so sorry, Adam, for putting you on the spot. But this is real today.

Everything that's here will be publicly for everyone in, say, a year from now or so. A whole bunch of people are productizing this. So this is just a little insight. Any questions on this?

Post-Demo Discussion and Questions

Yes, Mary. I think you're really bold.

So the demo is incredibly impressive, but it seems to me maybe it shows that the underlying LLMs still are giving pat answers that maybe aren't as insightful as they could be. Yeah, so the challenge here is I don't have a fear-based mindset and so the insight wasn't actually that good. The insight here was strongly created by the answers that I gave that wouldn't have been my actual answers and the fact that I specifically told it that we were in a demo

in order to take on a particular thing. That is because I do these demos a lot and you can imagine if I do a demo in front of some company and it starts pulling in data from my real emails and starts to actually talk about companies we work with, that is a problem for me. So that demo playbook specifically is set up to not be real.

I can tell you I've had many conversations, I mean this is crazy, with my executive coach playbook that have been fairly revealing. At least they felt to me.

Any other questions?

Sorry, can you, I'm not entirely sure I. So I think I can show this. Let me see, I'm going to, I think I can show this. Nope, that's not where it is. It's not showing at the moment, so it's going to take a bit too much time.

Future Applications and Use Cases

It's called MCP, Model Context Protocol, through which you can expose a whole bunch of apps to your AI that then can start to communicate with it. Each of those Model Context Protocol servers can have multiple actions. I've got about 12, 13 or so of those servers, and they each have, I guess, 20-odd actions.

So my actual use cases, the way I would use this every day, like I have at this point automated about 95% of my follow-up emails. Basically, it just takes a transcript. It looks into what is my general follow-up email. It creates proposals for me on the fly because I've got examples of proposals that I have. It knows exactly the product offering that we have. So it can go and go through that flow.

And I do prank phone calls.

One more question. Sorry you had a few times your hand up, so I'm gonna ask that yes So the question here is can you do the same with Microsoft tools and No.

The models underneath them are not as intelligent as the latest models that are available on OpenAI. No, they call them the same. They are not the same. I can guarantee you I've done live demos with all of them. They do not follow instructions as well by a long stretch. They don't have the same context windows. The other thing is they don't support the MCPs in the same way. So right now, you can't.

I wouldn't discard Microsoft by a long stretch. I think they might very well come out winning. Just right now, there's no way I would be able to do this within the Microsoft universe.

Closing Remarks

Now, I'm gonna be around, obviously, for the rest of the evening, so you can come and talk to me about it. In the meantime, we're gonna have Simon follow up. Thank you very much.