The other one anyway.

Oh, beautiful.

Excellent.

There's a little 5 second time delay getting underway there.

All right.

So the big episode 23.

23 Yeah, I mean, I missed last week's, but it was very good.

I enjoyed that.

That was a.

You got to be a.

Little chat with the with the big man himself.

It was good, yeah.

It's I'm still like slightly annoyed but it was on episode 22 when clearly it was destined to be episode 21 but it just the timing didn't.

It really was.

What can you do?

What can you do?

It was good.

I think you guys like covered some some great territory.

I think listeners to this episode probably going to be a little disappointed there's no ambient lawn mowing in the background.

That was a feature of of last week's episode.

I quite enjoyed that.

So I've got the doors open though, just in case anyone wants to start lawn mowing outside at this time.

Similarly.

Just in case.

Big big window open behind me here, trying to get a bit of a breeze.

So you never know.

Let's see what happens.

I thought you guys covered some good territory.

I quite liked, especially, like the second-half of the episode where you started getting into just, you know, building businesses.

And I thought that was.

Yeah, that was good.

I thought it was well received as well.

Yeah, Well, I mean, it's, it's a hard one, right?

Because we're, you see, there's a lot of focus on things like nostril over here at Sovereign Engineering.

And you know, we have a whole company called of stuff and it's kind of baked into the pie a little bit.

But like that's where we're going to end up.

But also we don't focus on it to start with because it's, it's true.

It's not, I think it's, it's a good way to build stuff.

It's going to be a very like empowering technology for individuals.

But it's, I'd say right now it's it's just like a little bit esoteric to use as a sales proposition.

Most people tend to get the the value of AI.

But it was good.

It was, I thought it was an interesting conversation, but like, honestly, like just walking around like in the blazing sun at lunchtime, recording it with traffic going.

Past you and stuff like yeah, yeah.

Well, you get it like you do get into the flow like and it's, it works.

But it was Yeah, yeah, there's.

Quite enjoyed how you guys could be mid conversation and then you'd have to break into a game of real life Frogger trying to cross the street.

It's just all sorts of, I think, like knowing which way to look.

Yeah.

Yeah.

It sounded hectic, yeah.

Yeah, it was much more, much more laid back today.

I'm just in a chair, my desk, so that works.

Likewise.

Yeah, excellent.

Well, yeah, I'm glad.

I'm glad you guys pumped that episode out.

That was that was enjoyable.

So I've got a few more clips to share of that episode as well.

So I'll probably just like slow release those over the next week or two.

Just spread out, spread out the content all.

Right.

Well, let's let's pretend we're going to record a podcast today as well.

So we.

Should probably.

So maybe like I don't yeah, I've never really set the last week right.

Like I was being all cryptic about where I was, I was doing and then did like a whole episode on sovereign engineering.

So but so I'm over on the the beautiful tropical island of Madeira, which apparently is having much better weather than Perth right now.

So I've I've won in that in that case, but.

Until next week.

Next week, I'll be sat here just in full linen.

It's going to be great.

Is it pretty hot?

Is it turning?

Is that?

Yeah.

OK.

Well, that's good.

Oh, it's like just no chill in Perth.

Perth, it goes from like wet and miserable to stinking hot essentially overnight.

So that's what will happen next week.

There you go.

We got the weather chat out of the way though.

Madeira, give us some updates.

What's happening over there?

It's nice fixing, fixing the Internet and AI and social media networking.

So it's been interesting, right?

There's been a lot of, we're focusing a lot on like how we can do more stuff away from computers.

You know, it's like different ways to architect different like Internet applications and services.

So they're more local 1st and things like this.

So, you know, you carry around like to be like a little super computer in your pocket at all times.

You know, like why, why is it effectively a dumb brick that can only talk to Mark Zuckerberg's computer?

Like that doesn't make any sense.

Like we can do all sorts on this a little device.

So a lot of people looking at, you know, how can you bring more, you know, like local data storage to the device?

Could you make less engaging forms of social media?

You know, you don't feel compelled to look at them all day and you just get like summaries of the day or like newspaper app or some all these things, different ways of routing traffic.

So you don't necessarily need like IP addresses or fixed IPS or this sort of stuff, which would be very impactful.

So it's been, it's, it's very like interesting creative space.

It's like, I mean, I, I launched a new update.

Well, one of my proposals to Nosta, she's like Nosta KV Connect, which is a kind of bring your own database to the app system.

So, you know, instead of having all of your data publicly published on Nosta or sat in your browser, you can just bring the database that you have at home into the application.

So you can have one common application with different back ends for each person, which is quite useful for, you know, maintaining secure information and things.

So you can have like a a Noster app for your company discussion boards was one of the the pilots I did last week at demo day, where instead of the information being on Noster, you could have a private version.

So you plug in a code and all of a sudden an extra set of boards just pop up and that could be your company boards.

And then you can interact with them with the same identity and everything else and all the security controls, but the data never touches Nosta, so you don't really see any other stuff.

So way more secure and stuff like that.

So yeah, I also did like a.

That's cool.

Like a wall, like let let me, let me interrupt you though, like what's your like?

What's your feeling about like business seats, applications for like on Nosta?

It's like, do you think this is the kind of thing that will get traction with businesses?

Because I do wonder sometimes if there's just like too much like complexity baked into getting started on Nostra, and if that's like a reason not to, not to engage with it.

I think right now Nostra is a, it's seen by most people as just like a social network.

And there's not, if you're just there for the social network, unless you're in one of the subgroups where there's a lot of people on, on Noster.

So that would be bitcoiners, people that like Noster and weirdly the Japanese like tons of Japanese people on Noster.

And you know, it's, you might not get what you want out of the social network there.

But the the interesting thing is that Noster is some first and foremost, really it's like an identity that you own.

So that's nice on social networking because you own all of your posts and you can verifiably show that it was you that posted and whatnot.

And you can take your identity across many different apps and it's portable and owned by you.

And that's all great stuff.

But it's not necessarily a big problem for like, I think day-to-day people.

So I don't know if we'll necessarily get them with that.

Now the interesting thing is that when you take that identity and you go and log into a new app and that app now already knows who all of your friends are and all of your followers.

And it already has like a wallet that's associated with you because you have money on monster as well.

And it also automatically has, you know, like a history of things that you've been doing or different services that you've paid for previously.

Like that's actually a better experience.

Almost beyond the social network is, you know, if you're, if you're, you know, you've spent a lot of time developing a particular network of people and things that you, you know, you work with, it's, it's useful to be able to move that around and to have that everywhere.

So I think longer term, what we, what you will have is a lot more applications that happen to use Nosta for kind of like the authentication side of things and for the back end and for managing some of the data in a way that is less secure.

I mean, what, what you see a lot with a good pragmatic example from why you might want to do this as a business is something like identity theft and securing people's data.

Like it's, it's, it's kind of a joke.

Like this is only really a problem because we kind of force so much data to be collected in the 1st place.

But, and then, so you, you force like Facebook or somebody to have everybody's data and then say, like, Oh, I can't believe you've lost this data.

And it's like, well, if you put all the data in one place, then you know, every hacker that wants to exploit that data is going to come after that data And they'll, they'll probably find it, you know, like it's, it's going to be hard to do.

If you create a massive honey pot, you know, all the bees are going to come.

So the actual solution to this is you don't put all the data in one place.

You spread it about the place and you have it encrypted and you have it under like the ownership of the individual.

And then now instead of hacking like one database in Facebook, I have to hack, you know, like 4 billion individual databases that spread around the world.

It's, it's just uneconomical.

So you make the defense that much harder, but it's so, so I think people will start to go on to this and realize that you know what?

I don't want all the exposure and risk of running all of your data and managing it.

I'd rather just build services that you can pay for very easily because you have money baked into this sort of protocol.

But at the start, and then you just use them.

I won't store any data on you because I don't need to.

And then I, I get rid of all the liability, all the risk and everything else.

And you can just take care of that yourself.

So there is like with that comes a little bit of like self responsibility on the user side.

So you know, it's based on public private key cryptography.

So you are managing like a cryptographic identity, which is a key, which sounds odd, right?

But then, you know, there's ways to do this.

There's programs you can use, but really what we're talking about is saying that you've got a secret big password.

And if you say it like that, people are OK.

Well, I, you know, I deal with passwords.

That's fine.

Yeah.

So it's achievable and it's all going to get made easier.

And I think people get more used to it.

But I think the benefit.

I think that's true, yeah, Is.

You know, like you minimize your liability, you can protect your own data, you can verify what you're doing, you can take one identity around on the Internet in a way that you know is useful to you and under your control.

I think that's going to become more and more useful the longer things go on.

And then people will just develop apps for this.

Like it's annoying to have different apps to sign into, I don't know, Slack and something else.

Like if you just have, like you say it's me, here's my identity and you bring it to the app.

And then the app can go and get what it needs from your public records or from your private records, depending on what access you get at.

Yeah, that's true.

It it almost sounds like simpler when you describe it that way, like from a user perspective as well.

But it is kind of fascinating how, like vernacular tends to shape a lot of this stuff.

Like just the mere fact that you might refer to something as a private key, then all of a sudden it's like, oh, what, What is that exactly?

Like I have to try and like conceive of what this thing is in my own mind and like, what risk does that create having to look after or hold on to that?

And you know, like the the entire animus shifts like the moment you just describe it as a password.

Yeah, I mean, and I have like a bunch of sub accounts that I, I manage like inner password, I'm trying to stream them like that.

They're not like knowing my knowing my main identity, I'm a little bit, little bit better at managing that one.

But it's, it's OK.

I mean, again, like the fact that these identities and these keys aren't in like 1 place to be hacked.

Like it's like, yeah, people can like there'll be stuff that goes out there at some point and tries to hack an Australian entities, but you, just you, there's ways we can revoke it.

Yeah, sometimes I wonder though.

Like the decision making process for a large company is usually just plausible.

Deniability is like, is there a big third party we can pin this on if it fucks up, It's like you almost don't need to worry about the security privacy benefit.

It's just like, oh, is there someone else that we can just handball this to if it if it if everything goes like balls up.

And so I just, I wonder if that's part of the.

I mean in any.

Companies look at this.

In any Australian company you have to have like a chief information security officer, one of the roles as the chief information security officer is to go to jail when you have debate breach.

And so you know, someone's asked me like, oh, do you want to be there for mine?

No, no, no, not just.

I'm not interested in that.

I'll do without that and so.

You know, I think under those circumstances, you start to say like, you know what, I would prefer if we just didn't have this data.

If if I go to jail because this data has been leaked, the correct thing to do is to not have this data.

Like, and this is the nice thing about monsters.

It's very like natural in the way it's been built in the sense that it's like, yeah, that's it's my data.

It's not your data like I do.

I have to give it to you.

Well, maybe I can just give you the bits that you need or something like that.

So I can just give you enough access or I can just prove who I am without necessarily giving up all the goods of exactly what I'm doing.

So it's very, it's very useful, like when you start trying to build apps with this stuff and you realize, like, oh, I don't need to use, I don't need to implement any identity authorization systems into this.

Like I don't, I don't even need to need your key.

Your key can live somewhere else.

You just give me a code and I can ask that person to sign this data and then off we go.

Like it's, it's very powerful.

And I, I think there's gonna be some really good example applications coming out of this.

But there's also a lot of just like it gives you a very nice like overlay network for a lot of different things as well.

So, you know, I've been working this guy, Jesus, he's made, it's gonna be called context VM that he's got out there, which is just, it's like magic.

And I definitely talked about this before.

Is that like I was, I had like a hacky version of something similar called Nostril MQ.

But you know, I what I really wanted was a legitimate software engineer to build something similar and he's kind of done it.

And this does like it takes the MCP protocol and then uses nostrils to transport layer.

So, and one of the benefits of nostrils, I don't need to expose stuff on the Internet.

I can just have web sockets from a computer listening to a relay and I can get the traffic in an app.

So it means I could have like a bunch of like compute services on my machine at home.

And I can have like a web app that's just almost like a flat HTML file and it can now act as though it is a complex web server from the HTML file because it just talks to the things on my home, home PC.

So you can imagine like, you know, you can have like a single interface that exists on the Internet and it changes completely depending on like what context VMS that you bring of your own into that party.

So I mean, I'll be demoing some stuff around that in the next couple of weeks.

Like I'm, I'm a big fan of this idea of like, what's the minimal amount of app that I need to run to make something useful?

And then you as a user can just, you have like a little box at home and you get a little App Store on there.

And you say like, oh, I want one of these and one of these, and then that completely changes the interface of the think that is available on the screen.

So like that's that, I think will be like a nice, nice paradigm.

And again, it just keeps like, why not just have the data in your house?

Like it's the same same way to use windows.

It's like, OK, well, I could do, I'll just have a thing in my house that just works for me like non-stop.

Like do I really want everything I'm talking with a IS to be out in the world?

Or would I prefer to have like a little box in my house that has a GPU in it that can run local models and can do things on my behalf that are private?

Like that seems, I think there's like massively feeling to people at the moment as they realize, you know, the same guy that sells you IAI is the same guy that tried to scan 8 billion people's eyeballs.

Like it's like, you know, if you want to sometimes they just tell you who they are and you're just like, oh, right, you want to you want to scan my eyeballs, Like just eyeballs or balls, balls as well.

Like how?

How invasive do we need to go anything like it's?

Yeah, no, that's, I think you're right.

I think everyone's starting to figure, well, not everyone, obviously we live in a bubble, but people are starting to figure out that, you know, you want AI on its own computer and then you want to be able to manage a squadron of, of agents and, and they just keep solving these tasks and give you daily updates.

And it's like having a little employee that lives on a.

Well, lives on.

Mac Mini in your case.

Yeah.

So this has been my my goal this week was, so every week here we're going to do a new demo.

So you've got to effectively invent something new from scratch every week and build it.

That's part of the contract, which is which is a thing and I think the idea was that we can't really do that because we have, you know, vibe coding these days and you know, it only needs to be a demo.

It doesn't need to be good.

So like, what can we do?

So I've done wallets, I've done the discussion boards, I've done like the Nosta KV connect.

I made like a multi person betting game with no back end to it, which is like it's hard to pull off in a single HTML file.

It doesn't like work usually, but we got it.

We got it to work and then, then we had, yeah, this week I decided what I'm going to do is I'm going to build an entire demo from my phone.

And so I've been spending the week using Wingman to do all of my code updates.

And it, it really came down to like, like this, this just like completely across the board.

Like everybody last week was just like fucking hate code.

He's just like got off on one.

And it's just like, I don't know, like 2 weeks ago or three weeks ago code code was like a genius savant.

And then last week he just became like a dribbling idiot.

And like, it just, yeah, it's just, it's constantly like building stuff you don't want or going like, yeah, what it did right is I wrote this test where I just didn't do any of the back end.

I just made it all up and you're like, oh, oh, that explains why it's not updated directly then.

It's because it doesn't exist.

And it's like, yeah, yeah, it's good and it works.

Success.

Like I've done.

Everything and you're like, no, that's not a success.

That's that's the opposite of success.

What the fuck are you doing?

So I got so annoyed at like, I mean, the wallet that I built last week, I built it 6 times and like, I eventually put it over the line by like having to like be quite deliberate and how I built it.

And I thought, you know what, what I'm, what I'm finding here is that I, this tool was good enough that it made me very lazy.

And then what I need to do is go back to that like slow code methodology.

And I thought, well, I wonder if I can do the slow code approach, but I do it from my phone.

So I do pilot.

So I like do segregated pilots to different parts of the functionality, make sure they're working.

And I take those and then I write up like a script for how to implement it from scratch.

And I just very slowly, like piece together the Lego blocks.

And it's been working quite well.

I mean, there's definitely been moments where I've had to then.

So I'm using my laptop is available for like fixing wing man because I can't do something or getting myself out of a jam.

Or something like that.

And then some of the context VM stuff I set up like using the laptop and that's providing some back end jobs for me now.

But it's yeah.

But like the full like the website for the demo will be from the phone using a lot of like dictation, like letting Wingman do his thing and come out with plans and implementation.

I was trying goose sub agents for a while and so I have some like recipes that then have like sub recipes.

So you have like an orchestrator and he has no access to write code, although he found a way around that.

Come on.

But he should be like reading code and dictating plans and a developer that does the the implementation and then like an architect that you can talk to and attest there and another agent who could manage the like the back end terminal sessions.

So you can spin up new sessions and then start and restart different, different apps for me, because like doing that sort of DevOps stuff is just, that's hard from the phone.

Some of the testing is just a bit of a pain in the arse because you just it's just way easier to test stuff on a screen than it is finalize start stuff up and you know, not your browser in on a phone doesn't really give you the same like console outputs and stuff checking what's going on and the error messages, but well, so far.

It's been pretty good, feels like I have there's a time in place for it now.

Yeah, so I have this.

So the goal is building like a bot called Craig David.

And like this is this joke is going to fall so flat because like nobody gets the the background over.

It's like.

Since.

Like international audience, but it's going to be so the the you know, the concept is, you know, in the year 2000, Craig David had a great seven days.

You know how does yours compare so.

So you.

Go to the Craig David bot and he'll go and get all of your events from Nosta for the last seven days.

Display your week.

So that's all working and then the next thing you'll do is that you can pay like a cashew token and he will go away and do some research.

And so it'll take all your seven days and it'll give you like a funny summary for every day and then a summary of the week of life.

So how did your week compared to another comparison of like your week to create David's week?

Who had the better week?

And then eventually the next step after that is going to be can we get it to write to take all the because we've got all the media that's available from your nostril events as well.

So that's like even all your pictures go into the the thing as well when it creates the summary, so you can see what you've been up to and where you've been and all that.

And then the final bit is going to be, can I then use like a goose agent second to edit a video on the fly from your video uploads and imagery that will that I can then set to the seven day soundtrack and then have like a custom wrap overlay.

Like, I don't know if I'm going to be able to get that, but that's the goal.

Nice.

I like it.

Yeah, I like it.

I like the reference.

That's pure nostalgia.

Yeah.

Yeah.

And so I've got the it's it's kicking around online now.

It's got the little both selector Craig David face so.

I need to, yeah, I need to jump off.

I was having a torrid time on Nosta yesterday.

I don't know why.

Everything was a bit buggy.

So I just ended up leaving it like I was.

I was getting a bunch of well, I put AI, put a note out just saying like who's building cool shit with AI on.

Not sure I want to follow you.

And I got some good, some good suggestions in there, but I went to follow like a bunch of people and just outright refused to let me follow.

Yeah, a bunch of people.

On both.

My browser and also on the phone.

So I was just like, I'm just going to leave it.

I'll come back to it tomorrow.

I mean, if you were you on like Primal on your phone.

I was on Primal, yeah.

Yeah.

So if you go, if you go to they, they have that like little discovery packs if you go to the Discover bed and those discovery packs, if you go to like the Sovereign Engineering discovery pack, it's like though that's that's a good one.

OK, I will do that.

People that are building monster stuff.

Yeah.

You know, sometimes it's just about getting some engagement on on.

Yeah, you just need to follow more people, post more stuff.

It comes eventually.

That was brilliant.

Yeah.

I knew a few people would respond to it, so I was just like, OK, cool, let's let's get people's suggestions.

But that's cool.

I like the sound of that.

I like the sound of that a lot.

So have you have you noticed any improvements in Claude code like in the last couple of days?

So obviously Anthropic put out a statement about this so.

Yeah, 'cause it is.

I've heard this from people like all around the world, like everybody of years has struggled a bit in the last week or so.

So it feels like they definitely nerfed it.

Like it gives you that sort of vibe of are they about to release like a major update or something?

Like because it feels like morals always seem to go shit just before before a new.

One comes out.

But because I've been using this week, like my focus was on like, all right, I need, I'm going to use Wingman to deliver this thing.

So like usually I would have like wingman for the planning and then I just clawed for the, the doing and like what I thought, right, well, I'll take the opportunity while I'm pissed off at claw code to just actually put the work into making better recipes for wingman so it can control the the code updates as well.

And it's definitely a different experience because you don't, you see, I mean, you don't even really see in core code exactly what it's up to most of the time.

Like when you in wing man, like it's the goose, like it's, it's just hidden.

Like it's like it's really, it's really awkward to kind of see what's going on.

But it's yeah.

So I I haven't really seen too much improvements in core code because I've just not really been using it all now on the same basis that I would have been last week.

But I don't know how it's your experience.

It's, it's interesting because I, I mean, I've obviously just been like slow coding, you know, I try to like be really diligent with that.

And yeah, the output, especially in the last week, it's like just got this inability to follow the plan, whereas in the past it would be really meticulous about following the plan.

The whole point of like having these design docs is really just to give colloid code scaffolding, you know, to be able to navigate through a road map.

And and so that's, that's the thinking behind it.

And it's kind of been amazing the last week and a half, actually, probably just the last seven days, to be honest, how often it will just ignore like huge chunks of the road map, skip steps, build stuff that isn't meant to be there.

And so like, I mean, I think I mentioned I've still got my cursor subscription.

So I just like pull that open and I'll use GBT 5 or groc 4 as a, as a review model to just go through and go, hey, like, what parts have we done?

Which parts did we do the appear broken?

Like where have we just created a bunch of like hard coding?

Like what what is going on here?

Because there's been just breakdowns everywhere.

So that, yeah, I felt like about 7 days ago, it really started to get like super derpy.

And but every so often they'd be like, I don't know, they'd be like, it was like prone to bouts of genius, you know, like it was like just utterly derpy.

And then every so often it would surprise the hell out of me.

And I was like, Oh yeah, this is, this is the old chord.

Great.

And then like 5 minutes later back to full on derp mode.

And so that's been, that's been the experience this week.

And I've just like been smashing through using cursor as well and trying to get different models.

I honestly think like in the future, this would just be something we do.

We'll just switch between different models, different days and tasks and, you know, we'll become more accustomed to it.

I think what's surprising about it is like cloud code is like almost had a monopoly on code creation now for ages.

Yeah.

And now all of a sudden now everyone's switching to codecs.

People are loving codecs.

I haven't really.

Used it yeah.

A lot of people who moved to codecs last week just to get stuff done to fix yeah it's like as soon as there's a mistake like this in there, so you see how quickly everyone is all right, I'll just try follow up the tools and see which one works.

And then you want that moment of switch.

But until then everyone's like, I just use this because it it works and it's there and I and it's not clear like where it was that they they broke it like was it like, are you using like a different Dumber model in the back end to save money?

Have you adjusted the system prompt in some way so that it's?

You know, attempting to came out done.

They came out earlier today with like a bunch of like basically just said it was like 3 interlocking bugs that have been like persistent for a while.

I think the one I can recall was like essentially a bunch of requests would like get rerouted to or misrouted to like to, to serve as with the wrong token limits and stuff like that.

And then that would that would cause a problem.

I do, I mean, there is like a broader question here around like how much trust you place in a lot of these companies and like whether or not they are like genuinely being transparent here or if like the the actions are a little bit like nefarious.

Because I think you're ultimately, I think like this is all just going to come down to unit economics.

Like at some point they have to make some money.

And it seems to me like the way you would try and do this is to like, even if you just say, oh, we're not going to degrade the model, we're just going to make it so that like the tooling essentially does get degraded over time because we reduce the token use like to save money and.

I think your mic is going to catch on your jumper a little bit by the.

Way.

How is it OK?

Yeah, I'm going to get in there, I think.

A little bit of feedback there.

Well, just that we are, I guess we are recording this one.

So I'll try try and be serious.

We are.

Yeah, the the, the ambience rubbing of the mic noises this week.

It's a little different from the lawn mower.

But yeah, I kind of do wonder, like if we if people just have more respect for companies if they just increase their pricing, just came out and said, hey, you know, we need to just keep increasing the prices because this is not sustainable in the same way that like, you know, like tech companies have always done this, right?

I mean, if you remember back to like when Uber first launched, like rides around town were pretty cheap.

And then over time things just gradually got more expensive to the point where like I, I don't like, I almost don't get into Ubers now because I just find them like prohibitively expensive.

And like I could literally revert to getting a taxi.

It'd be cheaper.

Obviously quality, not great.

I'm not going to do that either.

But you know what I mean, Just purely from a pricing perspective, it just feels amiss.

And it doesn't feel like, I don't know, they're like there seems to be like efforts to be transparent, but then like how believable is this stuff?

It doesn't feel very believable, especially when everyone is jumping ship as well.

You would think you'd be breaking your absolute backside to just be as transparent.

As possible.

I imagine a lot of this stuff is just trying to deal with inflation without admitting the inflation, you know what I mean?

Like it's everything's just got five times more expensive in the last five years because we printed like, you know, like five times the amount of money that was in the world.

So and then we, we try and pretend we haven't.

And things like anything in the tech sector particularly is, you know, like it's supposed the price is supposed to go down.

So like, it doesn't feel like your Ruby should be like 5 times more expensive.

But, but really the economics is there's a person in the machine and they're driving it around.

So it's the prices are going to have to go up because that person needs to eat food.

So, you know, like it's it's there.

And then I think there's probably a lot of not wanting to raise prices in a way that then like forces the degradation of the quality and the service and various things.

But as always, like just the government stealing money from you via inflation and pretending that they haven't is the root cause of like almost all the problems like this.

You know, if we just, if there wasn't by a small country of people in the world that are allowed to just, you know, every few years come up with an event that will steal like five times the amount of money now, you know, like it's a lot easier to get on with shit like if that doesn't happen, but you know, we have to have GFCS and various things and then all of a sudden everybody's surprised why they feel poorer.

Like, so we didn't even text you.

No, no, you just, you stole it in the other way that you use.

It's frustrating.

Yeah, exactly.

I mean, I, yeah, sometimes I do wonder if like an even simpler answer for Anthropic here is just bureaucracy starting to kick in.

I mean, this company is like scaled pretty hard.

And I wonder to what degree it's just all of a sudden we've got like, I think the head count's like getting up to 1000 people or something now.

It's like, how well do we actually manage that scale?

How much time are we just like having with people sat in meetings now?

Like, I mean, the absolute worst would be people now just going, oh, we need to think about AI safety.

And so we can't like, allow the model to do XYZ.

And so that means we need to stymie it in certain ways.

And I kind of just wonder if this is Occam's razor to some degree.

Like, is it just a bureaucratic base now that it wasn't six months ago, or it was just a little bit more manageable or, you know, there are fewer guardrails to progress six months ago?

Yeah, I mean, this is definitely one of the reasons I got like back into using Wingman and Goose this weekend was like, all right, well, yeah, like I don't, I don't think it's like immediately obvious to a lot of people is it's kind of like two things going on when you use code code like 1 is that there's the the actual model that it's talking to on the back end.

Like you don't know necessarily that that is the model that you think it is when you're using something like this where it's all baked in, but you also then have like the code that executes on your computer.

So like, you know, when the actual agent loop is something like it makes a request.

And so I want to achieve this and then the model on the back end knows that it can, it has a set of tools on your computer and if it sends back a certain format of message.

So instead of just saying that, Oh yeah, great, that's a great idea, it says tool use like use, I don't know, terminal to edit file or something like that.

And like you don't see any of those messages.

They just like come back in in the back end.

And so it's not necessarily clear like which bit of the the process is the bit that's getting changed here.

Is it like the way that they is it the way the app works on the computer and interacts with the computer that's changed?

Is it like the system prompts behind like the way they're using the models or is it models like one of the things that I think is kind of nice about Goose and why something like that, like whether it's Goose or open code or some other coding agent would be the one is going to be important.

Is that like, all right, well, I want, I want to know like what the system prompt is that's getting sent.

And I would like to know exactly which model I've chosen and whether that is the one that's running and potentially even have that local.

And then I would also like to know like what the the like the the tool calling loop is that is running on my machine.

And like, have you changed that?

Have you introduced new things?

And it's that tool calling loop, like it's very important.

But even from like a pragmatic like getting stuff done point of view, like with core code, like is that they just changed stuff last week and everybody I know became like 10X less productive.

And it's like, yeah, I like that shouldn't happen.

And you know, everybody that was using Goose was already 10X less productive.

So it's fine.

But you know, nothing changed for those people because they didn't update their settings.

And so it's very meaningful to think that, OK, well, where you know, who gets to update the settings on my workflow?

Is it me or is it just like some guy in San Francisco or somewhere?

Like it's like you want even if you can't like bring all these models, like internally there's a lot of like a sovereignty that you can get over that stack just by choosing your software differently.

But it's, yeah, it was, it was quite eye opening for me.

That was like, all right, I don't I don't want this variability of now Claude coats is going to decide to change stuff.

And so like I I think I'll probably just cancel like my Max subscription to Claude there and just go back to a standard 20 bucks and spot.

All right, well, I'll use it intermittently when I want to get something like I've got a peace problem that I want to get looked at, but I don't think I need to be charging you giving you like hundreds of dollars a month because I'm not just running this forever.

Like whereas previously it would be like, here's like just a screen full of Claude carnations, just like boost off minions kind of work out in the end where it's like, no, it's like, I think I need to go back to you need to be involved in the loop.

So this, I mean, this is a big difference this week.

And I actually know how this application works and it works how I've designed it, and I know it does because I built every component of it slowly and slowly in the way that it'll be like, all right, just give me a website, OK, now give me a website that displays this, OK, Right.

No, that didn't work.

No, why doesn't it work?

OK, OK, let's fix it.

That's how we get it to work right now.

Now that work.

OK, give me a button.

What does the button do?

Don't worry about that.

Just give me a button.

OK, we've got a button.

OK, right now when I click the button, I want to post to this endpoint that doesn't exist.

Don't worry about it.

Like let's do that.

OK, I can see that it's failing, right?

OK, now we're going to build that endpoint just.

And you step through it just slowly, but it's it's almost like that quick that like it's, yeah, it's not quite as quick as you can describe the system it comes into being.

But it like it's not far off.

But you can use it in this interactive way where you do slowly describe the system into being.

But I don't think you can just describe the end system and expect to get there like this is this still feels like the bit that's missing is you can help me plan.

You can come up with ideas, you can give me PO DS and all this stuff you apparently called open AI whoever what you can't take this plan and then slow yourself down and implement it.

And like we was the saying we've been using a lot over here is like load code is like a dog with two Dicks.

And that's the idea of like this thing just wants to code and you cannot stop it coding and it's just going to fucking go for it.

I love that you're, like ascribing that to the collective over there now.

Like, I feel like you've been using that for like a few months.

Yeah, we also introduced it's.

Definitely one of yours.

Like I feel like that's a That's one of your sayings for sure.

Yeah, I mean, we also introduced our We're not here to fuck spiders as a concept.

Good, good, good.

The cohort when I gave my one of my initial talks, which was obviously a reference to the idea that we should really focus on problems.

That need to be solved and just doing the minimal stuff.

And Gigi had a similar approach to saying, look, we we want full left curve solutions.

Yeah, you know, like there's the things that are going to work and the things that are going to get adopted are going to be the things that are so simple that they can just be put out there and they can be iterated on and they can work.

Don't don't mid curve it.

Don't over engineer stuff.

Don't don't make something like 10 times more complicated than it needs to be because that would be better.

Or I don't overdo the thing, just all left full group mode.

How do I do this in the minimal amount of stuff?

Do I need to like, can I just use everything that's already available?

So.

But yeah.

And the fucking spiders, it's just, you know, do the thing that's actually going to be meaningful to the world.

Don't, don't just create the next, you know, best way of doing a particular software implementation.

That's fine.

Like, you can fuck the spiders if you want to, but it's not what we're here for.

One of our three core values that add other stuff, yeah.

The best, the best values in the business I would say, but.

That's it, yeah.

Yeah.

Well, it's, it's good to get a little bit of Australiana over there in in Madeira.

It's good.

Yeah.

Representing.

It goes, it's gone down well, let's say it's very like international cohort like it's, it's interesting.

There's a lot of good ideas coming out.

So it'd be, I mean this, I mean, the next few weeks of demo days are going to be getting better and better as well because now we see people starting to build on each other's ideas and bring more stuff out.

So virtually cool.

Yeah, now that is cool.

Yeah, no, I like this idea of like not mid curving.

That's that's usually good advice.

I think that's just a good one to to follow in general.

Well, I think it's a tendency I've had in the past as well, which is the, you know, just design stuff correctly and not just move.

I know well back in 2019 or something like that, I think it was I was working on like E cash solutions for Bitcoin.

It's like an idea.

So I just thought, you know, I think this is where we're going to end up in the future and this is pretty Fenny mints and Ashu and all that and I I just thought, well, there's no way that anybody in Bitcoin would use this unless it is trust minimised.

So I need to find a way of like linking these E cash tokens to the underlying Bitcoin and some sort of claim where you can't just be stolen from unless and it's just like, I was like, I feel like there's going to be a way here, but I can't quite figure it out.

And then the fediment launched and it just didn't care and used like the federations to provide a slightly better trust model.

And I was like, oh, OK, that's interesting that that kind of does the job and turns out people are OK with it.

Like what was I doing all the extra work for trying to be clever.

And then they cashew came out.

They didn't even bother federating stuff.

It was just like, yeah, here it is.

And you're like, you know what, this is fine.

Like let's just let's just go to the left code solution here.

This is this is good enough.

Is it good enough to build a better global payment system than we have right now, with less trust and way more secure?

So maybe we just get a 10X improvement for the world as a minimum, and then we can always carry on later.

But let's just get it built.

I feel like the, the tendency to want to mid curve is a byproduct of thinking you need to follow a set of pre determined rules when you're starting a business.

And like, you know, I think we've all done this to some degree.

I've certainly done it in the past as well.

We like if I, if I need, if this is going to succeed, I need to go and raise X amount of capital in a in order to get X amount of capital, I need to prove that that ABC works.

And that just might be like stupidly over engineering the the solution to this problem, but we'll go and do it anyway.

I know.

And often like you do stuff like that at the expense of actually just going and building like the thing that works and getting some customers using it and just proving it out.

Yeah, there's a lot of a lot of a lot of mistakes just baked into repeating the dogma sometimes I think.

But yeah, that's the tendency now as well.

It's just to find like the lightweight version of whatever it is I'm trying to build.

Like I can have like a big idea in my head and it's like, OK, let me just deconstruct this down.

And there's not even like, I don't think that's like even like a minimum viable product approach to stuff.

It's just like, it's like, what is the optimal approach?

But like, does it need all of the bells and whistles?

Probably not.

Let me just strip all of that out.

I just needed to do this one thing end to end.

So I always thought of it as a bit like a, it's like a combination of a tall poppy syndrome and then like an ego issue for me where it was like on the one end, I want to prove that I'm being all clever.

So that's like the ego thing that's coming out.

Instead of just focusing on like the core problem getting solved, it's like, Oh no, no, I want to, I want to show everybody like how smart I am.

So there's a bit of that going on.

And then then there's also the tall poppy thing of like, oh, but I'm like, who am I to, to do this?

Like, so like, surely it's only going to like anyone's only going to take this seriously if I do it so well or so smart that there's this like magic piece that no one's thought of that that means that I'm allowed to do it.

And it's like, it turns out you're allowed to do whatever you want.

You can just, you can just do stuff like and I just should have done that for years.

Yeah.

So for true I mean I have to.

Learn all these lessons so late in life.

No, I mean, we're still young, Pete.

I don't.

I don't know what you're talking about.

I don't.

I don't always feel that.

Way we've we've got we've got time we're good yeah I know what you mean though like it it does feel like everything is just like a mad rush these days but I admittedly I've been like Claude coding obviously a lot less for a bunch of reasons, but like it was almost a bit nice the last few days because it's just changed switch priorities.

I've obviously been like spending a bit more time on some stuff for, for the business and some content and, you know, started building a, an ecommerce site because I was like, we need merch.

Like obviously we need to, you know, channel the vibes and the text making and everyone needs some of the merch.

So I was like, oh, this would be a cool project.

Like, and also we've kind of talked about web deployments and stuff like that in the past.

And I was like, how hard can it be?

You know, just straight out of the Jeremy Clarkson school of thought and, and just dive straight in And turns out that it's not that difficult, but you can actually make a bunch of errors, which I did make with, with cloud code that I'm now trying to fix with, with GBT 5.

So our fingers crossed we have a an e-commerce site coming soon, but that's, it's been nice to just do something a bit different.

We should, yeah, because we need a physical, we need a range of physical products.

Yeah, yeah.

We have hats and T-shirts.

I could do with some new T-shirts.

It's true.

Get your pre-orders in.

Yeah, that was interesting.

So it was good like it's, it's been, it's been good to just like, obviously like spending some time with some content as well just means spending more time thinking, which has been good.

And I've been looking at the rate of failure in AI pilots, which has been doing the rounds.

It's an MIT study that's been doing the rounds the last couple of weeks.

And that was, they were suggesting that it's, you know, upwards of 95% of AI pilots fail.

I mean, I assume it's a scoping issue, right?

Like it would be.

I that's my, that's my thesis.

That's not theirs.

No, they I wouldn't say that.

No, I, and I think a lot of the takes I've seen on this has been like somehow suggest, I know we've just been bagging the technology all episode, but I, I think a lot of people are framing it as a technology failure issue.

And like, I think my take on it has just broadly been, I know it's just a systems design and systems thinking issue.

Like it's actually just an implementation problem for the most part.

Yeah.

And really what you're trying to do is apply this old framework of digital transformation to a new technology.

And it's like what we'll do is we'll get more tooling and we'll just sling it at pre-existing processes that we've basically been been running for a couple of decades.

And we'll throw a new form factor of technology at this thing and hope that it works.

And I just don't think that's going to work.

I think like it's the wrong approach, but it also misses this like really massive opportunity to just go, how are we doing these things?

What could we be doing with AI?

And I think that is like such a massive opportunity.

But it also just reinforces in my mind that like a lot of people are just going to start these kind of challenger companies that are AI native from day one that just go and outcompete some of these bigger businesses while they're kind of digging around trying to figure out what the best way to implement the technology is.

I mean, having worked in a bunch of these bigger companies and a bunch of their like innovation style themes, like the way these things come around.

And so they go, we need an AI strategy and we need an AI pilot.

And like somebody that doesn't really understand that technology, but it's good at talking about stuff will come up with an idea.

It's probably nonsense like, and you could fail the project right there.

So you're right, OK, but you save all the money and just not do it.

This is just fucking nonsense.

Like you don't know what you're talking about.

The tools is not how you'd use this.

It's not going to make any sense.

And then you'll get a team that doesn't quite really understand the tools and doesn't really use them very much and they'll attempt to do it.

And then I don't know, then you'll like weigh it up a few years later.

You're like, well, what's the chance that this was ever set up for success in the 1st place?

Like you don't even need to think like you haven't identified a core problem and you don't understand enough about the systems and the capabilities to know, yeah, well, do the capabilities of the system map to that problem or not?

It's all generally very surface and it's done to just quickly spin up some projects and you'll have some like, you know, usually the guy there has got some mates and a consulting company and they're going to come in and they'll charge big fees and you know, it's only a few 100 grand anyway, so who cares?

And like, it's just, it's just money just getting pissed up against the wall for nothing.

And then you've got like the startup version, which is just, they just got all right.

This is the new theme spray and pray.

And like it's, it's just designed to be pissing money up against a wall from it's so frustrating because there's like so many, like good.

It's, it's one of the things that comes out being here is there's like a lot of very, very good, very smart, very principled people trying to build stuff that just don't have access to any funding.

And there's just millions being pissed away.

I'm just fucking nothing.

It's not going to help society at all.

And you know, I, I can't solve that right now.

So I'll just laugh at the absurdity of the whole thing and then, yeah, yeah, continue to hang around with all the all the widows who just do it for no money, just for the love of trying to make the world a better place.

You know, I do, I do think there's like a, like the other side of the, the other side of the coin on these pilot failures.

Is it, it got me thinking that I actually think like we need a, like a new, a new way to actually determine whether these things are successful.

Because I, I, I suspect that like a lot of the kind of traditional metrics for return on investment are going to be a little bit, a little bit broken, you know, because like there's an assumption that's baked into traditional return on investment that like assumes that value creation is additive, right?

Like we've talked about this with, with companies as well.

It's like, oh, we can save a bunch of time on a per task basis.

And therefore, like time's up by the number of time, number of tasks that you would, you would be doing with that process.

And then you can roughly work out what the cost saving is.

Right.

It's a classic, never realized benefit.

Exactly.

I think that the difference now is there's like kind of like a compounding value creation opportunity here where like I don't like each, each component like amplifies the others in a way that compounds if you bake in the technology correctly.

So like in that example, like with task, like automating like a particular task and you save a bunch of time.

Well, it's like, OK, well, there's a, there's the opportunity cost of where you'd then go and deploy that save time.

So what are you actually creating on the back end of that?

And how are you baking that into the return on investment calculus for how you automated that, that other process?

We you probably haven't because the benefits of the opportunity cost are going to be way downstream.

And then there's like, you know, I don't know what, what, what is that?

Like?

Let's let's use an example like the one, the one I was thinking of earlier today was like using AI for document review.

Like if you're a lawyer or an accountant or something like that, this seems like a very like straightforward use case.

It's like, OK, so you're going to get like patent recognition benefits over time, you get that's going to then lead into like risk assessments that are going to become like more efficient.

You're going to get like, I don't know if you're an accountant, maybe it leads to faster deal structuring and better terms.

Yeah, the ability to take on more complexity in your client files, which leads to new service offerings and more customers and market expansion.

How do you factor all of those things in your return on investment calculus?

Like I would argue you can't.

Like it's like how how do you account for the things you just don't that haven't yet emerged when you're actually doing that implementation?

So there's like a bunch of like downstream benefits and if you have this like timescale for working out when the return on investment needs to be realized by when you basically just kneecapping yourself because the expectation is like we've got to get this return within three to six months or like on based on some reporting, reporting cycle.

And it's like you're just not going to manifest those benefits that quickly because they're going to compound.

I mean, my yeah, that's pretty true.

I mean, it's, it's hard to know like how people are setting up these pilots and like how they're defining this.

There's been, I mean, I would usually do stuff like that.

I would be thinking that these, you know, these returns are going to appear over the course of like several years.

So we need some, you know, leading indicators to understand why are we actually hitting the things that you might get in three months.

But like, it's, I think more than anything, like I think we talked a bit about this offline as well, was that I just wonder if I'm, I'm mentally further down the track than I think most of society is.

There's like a mismatch between, you know, like what do we need with AI right now?

And it's like, I think because in the coding space, it's quite advanced because you've got the right sort of people coming together that we're all interested in tech anyway.

They're acting on it.

They're using it all day, but every day.

And then you have like outside of the coding space, people use AI.

It's become a thing.

It's there.

We can talk to computers, they can do searches and stuff.

But but for the most part, like, I think there's like a big difference between like, you know, what agents can do for you versus like your experience of using ChatGPT.

And I, I wonder if like, you know, we say like pilots.

So I don't know if we're really in the pilot stage right now.

Like IE I probably like 6 months ago, I would've said, yeah, yeah, like it's strategy and pilots, let's get going.

Like, we know how these things work, but that only works like if there's like enough knowledge in the market that people understand what the potential opportunities are with this thing in order to correctly formulate the pilots, where you can map your problems back into like, you know, things that AI can solve.

And like you can do it very like high level of going like, yeah, I'd love to be cheaper and useless people.

Yeah, would love that.

But then like, your ability to actually execute on that isn't that simple in this case.

But I, I really think that right now the thing that the world needs is like actually good training and like explanations of how these things work.

Like I actually feel like we're probably even a bit too early for potentially writing strategies on these things, never mind.

And I feel like, so I can sort of see that like it makes sense that the pilots would be bad right now because I think people are just misapplying the technology.

And I think like because even if you I.

Think you're going to say though, like sorry, yeah, no, I was going to say like I, I think the one thing though, like this is probably like the The Dirty secret is I think a lot of people inside of companies are using AI and they're probably just not admitting to it.

And I think that's probably going to be a significant percentage of people.

Like I don't want to guess at a number, but if I had to, I'd be like it's north of 70% of people are using AI.

So.

And that might just be ChatGPT to help them do aspects of their existing work, right.

And they're just not talking about it because they don't want people to know.

And I just think like, yeah, like you're right.

I agree.

I think there's like, you obviously want like better training and we're still in this chat modality, right?

It's like, how do we just implement chat bots inside companies?

I think that's like, I don't know, this feels like a really like horse and buggy approach to, to AI.

And I think we need to break out of that pretty quickly.

But if people don't know what we're what we're capable of just yet, then then they won't be thinking about it.

They'll just be thinking about the modality that's in front of me and that's where I think training is useful.

I almost think like just going in, like we've talked about like a speed run as a concept, just going through and like almost like taking like a mini version of what you guys are doing now, Sovereign Engineering and just running inside the enterprise and going like right here's awake.

Let's just go and build a bunch of stuff and you know, it's like pure creativity and experimentation and learning and see what comes out the other end of it.

It's like a bit of a hackathon but with these two on the tools essentially.

Yeah, there's I think there's a lot that you can do, but you need you need to use these things and you need to use them in the more sophisticated manners.

I think like this is this is definitely going to be a bit of a a development road map for me on wing man is like, how can I make a lot more of these?

Just what is a non coding agent tasks and workflows just more available and appealing.

It's people like it's because it is magic.

Like what these things would do.

You know, like I even I have my own, but I can do the deep research type loops that you would get in core code and open AI have that running from just wing man and but it'll build it out in like a a lovely, like a nice deep document store in Obsidian of like everything that's available to me on the island over here.

And it's like, ah, cool.

That's yeah, that's more useful than just having it inside like single report.

Like I don't need a single report all the time.

Sometimes I actually just want you to keep me updated on what's going on.

Like, yeah, it's I, I think that's, I think that's the right focus though.

Like I think it's hard to.

It's hard to get these pilots correct if the people that are going to then sign off on the pilot and what you want to do don't yet aren't bought in like it's so if you would have like this this board level like, Oh yeah, we need one of these.

We need ANAA strategy.

Like what's he going to do?

All right, we'll do a pilot and we'll figure it out.

And it's like, you're not going to figure anything out.

Like you need to like the pilot needs a real problem and a real opportunity.

And there's probably a preceding step here that you need to go through to, to just and like, again, this is these tools are really good for this because you can build shit quick and like it just just so I just wanted to call back.

I forgot about this.

But when you said like, you know, we've been ragging on the tools, it's like it is worth pointing out that it's like, you know, maybe I used to be like a productivity 1 and I went to a productivity 100.

Yeah.

And I've come back down to a productivity 90.

And my complaint is that I've come back down a little bit and I've had to pay attention.

It's like, it's like this is my complaint.

It's like, God, I have to like check in on this thing like every 5 minutes.

Like I'm every 5 minutes I have to do my job instead of every like half hour I have to do my job is the main complaint that I've got at the moment.

So I'm not, I can't do that, do that.

It's like I can't work for a full day without me checking without messing up.

It's like, yeah, most, most like junior people couldn't do that.

Just wasted.

They'll just do something else.

Think I'll do the thing they're interested in.

Maybe they are like sentient and it's just like but if I just fake this test then I can go and do something else.

Yeah, yeah.

More human than we think.

Yeah, yeah.

I, I mean, I, I think there's, I think you're right.

There's the, there's the, the animus for the return on investment.

I think that's the first problem.

It's like if you, if you're in a position where you need to somehow substantiate the return on investment within like a very short time scale, that problem #1 like I think you're setting yourself up for failure there right off the bat.

And then the second one is the availability of like expertise that can actually go and set this stuff up correctly and actually design these systems from scratch and figure out how to do that.

And I would argue that if you are particularly skilled at being able to do that and you're particularly proficient with the tools, why would you consult?

That would be, that would be the first question that pops into my head, mainly because, and maybe there's like a, that's a safety and comfort thing.

But like, if you're particularly proficient with a, why would you not build?

That just feels like that's where all leverage is and you're uniquely positioned to go and take advantage of it, take advantage of it.

So that that's, that's the next thing that necessitates that there's a lack of skills and expertise in the market to actually help you execute anyway.

And then the next one is just like these kinds of companies that were self reporting in the MIT study, they're big bureaucratic beasts.

These do not have cultures of experimentation, which is kind of what you need if you're going to run a pilot like it is by design and experiment.

We're going to try something completely different and we're using new technology and we're taking a big risk on this.

And so if you're going to do that, maybe consider there's like a Netflix incubation model, build the the workflow, build the part of the business off on the side away from your business as usual, get unique talent and to run it that isn't caught up in your in your day-to-day and go about it that way.

And yeah, that that might be the only way to to go and execute on this stuff effectively.

But yeah, I think I think ultimately it's going to be really difficult.

And the only thing I would say is like when you see headlines like that, just I'm filled with a lot of confidence because it just means like guys like us can just go and keep building stuff and the big incumbents are going to continue to do what they're doing.

And that's great.

That's great for us.

Yeah.

Like I feel like there was, you know, we've had the conversation tons of times about like what is the answer here?

Is it like a buy versus build thing?

And more and more I'm swing towards the build.

So I don't know if there's really people are focused on this or not enough.

Like I think you probably do have the time to build and then capture market share as opposed to needing to to buy these companies there makes more sense I.

Think it, I think we're also in like a really unique time where you've essentially got carte blanche on like where to focus because there's inefficiency everywhere.

You know, like it's, it's, it's really about like what problems you want to go and solve and like what problems are you actually drawn to that you want to solve?

Because that's the other thing.

It's like it's not where, you know, there's no, there's no upside to working on something that isn't really all of that.

It's not motivating to you.

It's not going to keep you up at night thinking about you're just going to, you know, kind of give up on this stuff pretty quickly because it doesn't, doesn't like you up in any way.

So there's an interesting has to be there.

There's an interesting Tim Ferriss podcast this week where he had this guy called Pavlos something over, but it's like an old hacker that's turned into like a venture capitalist sort of thing, like doing things with Blue Origin and all these like big bets somethings.

But it was but one of the comments he talks about was like, you know, like like when we think about that sort of like future looking investment and deep tech, like people think of Silicon Valley.

Silicon Valley has just been obsessed with SAS systems and software and stuff.

And it's like, it's been, this is like 2 to 3% of the, the world's economy is software.

Like these are very big businesses, but it's, it's not the money and it's all the money in the world.

Like, you know, the, the shipping industry is, you know, like 4 times as big and burns most of the money if it spends like every year just on fuel, you know, like, like energy and skipping and just becomes and all, there's all sorts of problems that needs to be resolved in the world that we just haven't really attacked in the same way.

And I, I did made me think about like, yeah, I think this is potentially where there is a big AI component.

It's not it's not necessarily that it's it's like a software play so much as it's a if we can get these things to do human tasks like it just like any industry, you can just take like a massive swing at and just do it way more efficiently.

There's the, you know, even without taking on the tech risk of just does this work or not?

We can just take off the, you know, like standard.

Yeah, like I do think there's just, but I'm starting to get a little bit conflicted as well.

Right on the the human at the edge versus bionic human type thing where it's, it makes sense to me that the end game is most likely human at the edge.

But maybe there's more of a path dependency for bionic human work than is immediately obvious.

And you might be better off.

Yeah, but it might make a lot.

But maybe my the point there is maybe I always figured the bionic human is definitely more achievable at the short term, but I'm starting to wonder if it is more of a dependency.

Like maybe you don't even get to the human at the edge stuff without going through the bionic human.

But it's like you start by replacing bits of the human and then your humans are very productive or so I mean, because this is sort of what is happening in the coding, right?

It's not People are still what we realize is there is a bit here that isn't isn't this doesn't seem to be automatable at the moment.

So it makes sense to just automate the bits that do and then keep the human and the human is very much in the middle.

But the humans are getting, you know, like 100, you know, you just we're building new things like inventing and then building new things part time every week currently.

Like it's like it's amazing, like what you can get done.

And this isn't just like, you know, shitting out a new website like it's like that.

You could do it like it probably does automate that whole thing.

Like I don't think I would pay for a website to be built again.

Like it's just like this is this is this is dead.

I could just create one from scratch in minutes.

Like this is not hot.

But but like, yeah, like I, I, it makes it makes me wonder if I was just maybe I've been a bit off on that in the past and it was there is a little bit more to be done with the, the bionic human stuff, but.

I think there's some truth to that.

I also think it's like very much like role dependent, business dependent as well.

So I think the framing we've used in the past is like you need to like identify the optimal human placement.

And so the optimal human placement in like a particular workflow might be that it needs to be more of a bionic human where they're sat with the tools.

You've got an individual sat with the tools and is able to just do 30 percent, 50% more than they were able to do previously.

And there may be aspects of that where they now no longer need to steer the tooling and they can go and sit off on the edge.

So I do think there's like an interrelationship here.

I don't necessarily think it's a binary choice between the two, but I do think that you're right, it's probably in the short to medium term and maybe that is like still 567 years.

Who knows where you've got an individual that is still at the terminal doing the bulk of the steering with a bunch of like AI agents that they were that they interoperate with.

But but I do think it's like very, you know, context specific.

So the challenge is in finding what's the ideal weighting for for bionic versus human at the edge.

And that's a call back to I don't know what episode it was.

Maybe it was like episode 2 or 3.

It's like an early episode.

We talked.

About maybe 3 or 4.

In a lot of detail, might be, yeah.

So if anyone's still listening, all two people go back to episode 3.

And if you are still listening, chances are pretty high that you listen to episode 2-3 anyway.

You should.

Like and subscribe.

Oh, very good.

There you go, you got it in.

There you go, you got to be professional.

Professional podcasters.

We should, we should like bake that into the intro because that's when the bulk of people are still listening.

You know, I think they.

They'll find it if they want to listen, they'll find it.

But.

That's solid as always.

Solid stuff as always.

I think we're like over an hour.

So I think we're yeah, we probably hit our our rant limits as you all had, to be honest.

There's no such thing as a rant limit with us, but there's an acceptable limit for everyone else's tolerance.

It's probably more accurate with.

I think we could have like a we could have a solid YouTube like open math thumbnail just called it's fucked.

I think that would work.

And then we get into like, oh, it's not it's all about wingman.

He's wingman.

Instead it's.

All about wingman.

Yeah, Subtly seeding wingman.

I like it.

That's the growth strategy right there.

All right, All right.

Well, let's call it.

That's the good stuff.

And then stop this one.

It's not stopping all right.

I mean I've stopped it like locally on the machine, but let's see if this this just doesn't.

I just hit the button and it doesn't do anything, just.

Going to continue.

Yeah, there we go.

I guess we'll just have to podcast forever.

But why isn't this?

Like I literally got a stop button, I press it and it does nothing.

I mean, we've just been like.

Don't have a buttons do anything like what the fuck?

Like none of these buttons right?

I'm going to hit the leave button and see if that works.

Just the leave button or?

No, I can't hit the.

Let me try hitting the leave.

Button doesn't do anything?

Oh, you've gone.