All right, let's get ourselves sell.

Welcome to the.

Very first balcony pods.

Balcony Pod This is it from an undisclosed location overlooking a man mowing apart.

Yeah, some ambient mowing noises.

The background it's.

The big It's the big Episode 17.

17 We are nearing ever closer to 20.

21 That probably is the big one, the.

Milestone.

When will that hit?

So it's four more weeks, Yeah, if that's when I'll be away.

Yeah, another undisclosed.

Location Fair but undisclosed, Yeah.

I feel like we have already disclosed on the pod, but it's OK it can remain undisclosed for this.

Undisclosed for now, yeah.

I don't know if we, yeah, we've definitely talked about the place.

I don't know if we've talked about that, that, yeah, maybe not me, but.

Future Future pods.

I think there'll be some good guest recordings while I'm there.

Oh yeah, and tend to take the mics and go for walks.

Very good.

I think it'd be good.

I'm excited for that.

Good.

Good for reach on the pods.

Maybe.

Essentially.

So people have kind of like.

I haven't.

Moved to Nosta.

I have noticed that YouTube has made some little algo changes recently.

Oh really?

Yeah.

And there's definitely like a little less organic reach on some of this stuff.

At least that's what I'm saying.

It could just be completely anomalous.

But I'm not convinced that we're the best people that hyping up our own stuff.

No.

But that's OK.

But like we, we do keep sitting down and talking.

So that's, that's good.

Look, I think that's the the goal is consistency over time, I think.

Yes, and if anyone actually listens and finds it entertaining and valuable and even better.

But we'll keep going.

Yes, all right.

We've got a bit of sun, we're on the pod today Which.

Is nice I.

Mean the last couple were we had the ambient rain, today we've got the ambient mowing noise and.

That the ambient sun you got to get your vitamin D.

Speaking of like hype though, I've been like trying to build a little hype man for the pods.

Little autonomous hype man.

Yep, a little segue.

I'm looking forward to having a demo of this afterwards.

But I mentioned it last week, I think I was like, as a kind of a rough idea that I've had just before we started, like just before we hit record.

I was like, it'd be cool if I could just like quickly vibe code up a set of agents that basically take our, our podcast transcripts every week and do some web search and just start essentially building up their own stream of consciousness based on the stuff that we put out into the world.

And, you know, they're just, they're tweeting all day because we're not particularly good at sitting there.

Tweeting all day?

Want to tweet all day?

Yeah, exactly.

I've actually got like, I noticed there's a segue.

I noticed the the other day, I've actually got quite a long list of saved tweets that I've just never hit post on.

Like I'll often just like type them out because I just want to get get something out of my head and then I just don't hit record.

Do I hit drafts?

Yeah, it's in drafts.

Yeah.

Really.

Yeah.

Got to use the right vernacular.

This demonstrates just how little I use, use Twitter.

But yeah, yeah, yeah.

I was like, this could be like a cool little like way for me to play around with Claude code because that was the other thing that I wanted to.

I dipped my toe in last week.

I spent a lot more time this week and it's great up until I hit my hit my limit, which is very often.

OK, I'm going to wait a few hours and that forces me outside to go which.

Plan is that the normal?

Plan.

Yeah, I'm still on the on.

The right to stop it to Max.

Yeah.

Exactly 200 or whatever like it's I think even though there's limits there higher, I mean I'm on the 100 at the moment and then.

100 seems to make sense.

I think it was more just I wasn't sure if I would use this longer term.

I just wanted to play with a smaller project, see if I would use it or if I would just skip to just recode and just jump over that altogether.

But I think it's useful in some respects.

I mean, I'm, I don't get as too far off track, but I am like I'm considering giving Claude code another crack.

Yeah, this month.

But particularly with the sub agents because I think that's the bit that I would have that like when I, I compared.

So I've been doing a lot of work this week working with Goose to try and get that set up as like a, just with it being an open source agent, but like it's moving quickly.

They're adding a lot of interesting features to it.

And I was like, OK, this feels like where I might end up.

I wonder if it's good enough for me to go there at the moment.

It didn't really deliver for me and what I was trying to do.

And part of the reason was it just didn't seem autonomous in the same way that I want from the code execution.

Like it was good for the planning side of the of the business, of the business of me talking to the thing.

So it's good at that.

But like with the code, it would just go like, yeah, So what I'm going to do is I'm going to do this, this and this and this.

And like, yeah, like that's we're cracking on and it just stops and sits there.

I'm like, I just have to keep saying, but you haven't done that.

And it's like, Oh yeah.

And then it does like one step and it goes, yeah, and I'll probably do this next.

I'm like, well, can you do that next?

And it's like, Oh yeah, I'll do it.

I was like, oh wow, this is, it's like it doesn't have that, I don't know, built in bias to just keep going in the same way that, you know, that was what like code code was when it first came out.

It was just, it's just going to keep running until it thinks it's done.

RU code is very similar, but I mean, particularly like the thing that RU Code did well was the boomerang tasks and orchestration engine that it had where it would say, take the task, break it into many smaller tasks, and then do one step at a time and just keep pushing until it was done.

And then flip into like the debug modes and things when they needed them.

And I think now they've got sub agents inside Cloud Code like that's going to give you that sort of feel to it.

So I'm kind of keen on that.

And Goose has also got sub agents in there like Alpha pipelines.

I could probably try them, but I don't know, I feel like it's probably just needs a bit more.

But but I did also vibe code.

I was sharing with you like at the end of last week, a little project called Wingman, which is a way.

So you run the Goose CLI and then control the CLI like via like a, yeah, this little Wingman app that sits over the top of it.

It'll set up new sessions for you, give them tasks and then you could monitor it from anywhere.

Because like as we're sat outside now my, my goal long term is like, I would like to be able to code these apps.

I do not want to be sat in front of my computer.

And part of the problem I've had is that I, I can, I've code these things if I sit in front of my computer, but when I'm not, they stop.

And that just seems like, well, I'm just wasting most of the time that exists in the day because like, particularly like kids around, there's only so much time you can.

Yeah, also sit undisturbed.

Like.

Ideas and just prompts will come to you at different points in the day and you want to be able to go and action them on the fly and let the let the machine do the hard work.

I do kind of feel like a good agent might be the one that just says maybe no more ideas, maybe just like finish off some of this that's.

That's the next thing we should.

Probably.

We should probably jump in.

Do you want to go back to the Hope Man?

Yeah, I'm interested to hear about it.

I don't.

What I would say is like the first thing I noticed was like, I obviously planned this out quite a bit 1st and got a bunch of design docs built out because I just wanted to get a good idea in my own mind what this would be, how it would function.

I was very like cost conscious with it as well.

Like, I didn't want this thing to basically just go off the reservation and we'd wake up one day and we're spending hundreds of dollars.

It's not a big deal.

But it was more just an exercise for me to figure out how to manage this stuff in production.

So, yeah, as I said, like for anyone who's listened to this before, I basically use all of these projects first and foremost as a bit of a learning exercise.

And I generally have an area that I'll try and focus on to, to to try and spec up my own, my own skills.

And this was an area that was like, OK, well, I'm probably wanting to get at least some cost reporting built out so I can just see what's happening when it is ingesting in transcripts, doing web search, figuring out what it wants to tweet out.

And so I spent a bit of time on that.

What was interesting is glowed code does just go off the reservation.

Sometimes it will just go off on one.

And to be honest, I kind of just like ignore the, the, the natural brakes and just let it go.

And then I'll come back to it and I'll be like, oh, you've done like full cost report breakdowns based on like different model usage.

But I'm running everything through open router.

So it's not really necessary here.

This feels like a redundancy.

So I'm like having to go back and just be like, Hey, can you just remove all of this like cost comparison stuff that you've done?

Let's just have like a very simple cost report based on like the tweets that are that are going out per transcript.

And yeah, I was it, it just made these like decisions to just go and build more and more and more and more.

It just wasn't necessary.

And so eventually I just pulled it back and was like, OK, just do it like code review and just look for inefficiency here and what code is necessary?

What code is like just unnecessary.

And let's just strip this back so it's like most usable.

How did that work?

Because this is, this is one of those areas that I'm keen to automate and just have run every like overnight.

Like I feel like there's a lot of tidy up that can be done.

Down hours, but I must admit I cheated on this a little bit, so I actually went back.

So once I'd popped that into GitHub, I went back to called desktop to do this first because I wanted to.

That is, you know, I don't know that there's any material difference there.

To be perfectly honest, there might not be.

Between called code.

And yeah, but I just wanted to get in a different environment.

Yeah, there's a mental difference.

Yeah, nothing else, right.

Like it's.

It's just for my own like Project Hygiene, I think to just get out of the the coding space and into a different environment to actually think this through.

There are as well, like I mean cloud code is a coding agent that will have it.

It will have a different set of system prompts behind it.

Yeah, called desktop.

Like it definitely does behave in a different way, like both code stuff like that's fine, but like the way that they like exhibit is is meaningfully different I would say.

Yeah, well, there's also like a difference of what I'm starting to be able to spot and then using Core desktop as a way to essentially like triage potential problems.

And then I'm reading that and going, OK, I've got to be aware of the stuff in future as well.

Like to be honest, I missed a bunch of like get ignore stuff that just like API stuff should have been popped in there and but I was able to spot that ahead of time.

So I was like, OK, well, that's, that's OK, I can deal with this.

But getting into Cloud Desktop was good because then it just essentially prepared like a full code review for me.

And I read through that.

And then I was like, this is actually quite handy.

And then I basically just asked the same question back to Cloud Code and just let it fix most of it.

And then I went back to Cloud Desktop to do like a review of the updated code base and then do it a new, like a fresh code review.

And it was like, yeah, like 99% of all the stuff I had mentioned was was actioned.

Yeah.

And it made a recommendation for one thing but said it's like low priority.

Probably doesn't need to be done but.

Yeah, I.

Mean it was interesting.

I know definitely the way I have the cloud desktop set up, I think you've replicated pretty similar is that it can read the code files, but it's read only and that just makes a huge difference.

Yeah, it's like anything like this is like try and do like a code review or a plan or anything in root code.

It just, it just goes for it.

Yeah.

But it's like we said before, it's the dog with two Dicks.

It just can't help itself.

Like it's just, yeah.

Yeah, that was my only frustration with it is I found like I had a very good, very clear set of documentation around what needed to be built, how to build it out.

And I think for the most part it was followed.

But yeah, every so often it would just be like, I'm just going to take a little bit of creative liberty here.

I'm going to build out this thing that has just not been covered, not asked for.

So I, I mean, I do think that this is where this is the the beauty of the Ruco down orchestra mode to me is that like any, the large context of what you're trying to build is known by 1 recession, but then the actual execution can't happen in that session.

Like that.

That role doesn't have any of the tools available to it to change the code, but it does have the ability to write tasks for another session.

Yeah.

And so it does just write like a much, much like looser, like a much more tightly specified subtask for that level below it to complete.

And it doesn't know anything else about the greater mission.

Yeah, it will just look up on the flag.

Here's all the information you need to know to do just this.

And it doesn't know that it's doing anything else.

So it does just action that.

And I imagine they've probably spent quite a bit of time just raining it in.

Yeah.

Whereas I guess when you even if you've got like a plan, like if the whole plan goes into memory and it starts coding, it's always aware of the whole plan.

Yes.

And so it, it kind of goes, oh, it doesn't go over here like you don't know, like where it's placing the emphasis on something like that when there's just so much that you've got in your head.

I mean that kind of mimics like.

Humans working?

With a developer like this is what they do is yeah, I used to joke on me, I don't know, like 20 years ago.

It would be right in Gantt charts.

Here's our here's our picture.

This is basically like a picture of what we think might happen and it will just be nonsense.

Like as soon as we're done with this, even though we've sat down, we've planned out the whole thing, we've gone through this like, whole process to figure out what we need to do 1st.

And I said, I swear, like within a day, you're just going to go and do this thing over here that you've confidently told me you couldn't do until this 20 things are done.

And like, that will be the first thing you deliver.

It's just like, it's always the way they go.

Oh, yeah.

But I just fancied doing that, so I did it and it was OK.

It's like, yeah, that's always the way.

It's like you get distracted by something.

I want to do that one though, so.

There's there's definitely like a mirroring of like human experience.

Yeah, I must say, like that's something that I found that it's almost like a management experience more than a coding experience for me.

Like I'm obviously trying to use it to learn a lot.

But obviously we also want to, you know, push, push things out to commercialize them.

Yeah, it's not all fun and games we.

Should we should really do that?

That's the intent.

But I do find my coding experience is more like a management exercise and I've done a lot of that.

I've managed people for years and years and years, and that's just native to my understanding, whereas code is still quite abstract and but I find I've just.

I've naturally found myself gravitating to running these like I would a human project, and it might not be the most optimized or the most efficient or the fastest approach, but I generally get the output.

It probably is the best it could be.

It's pretty like it wouldn't, it's not necessarily going to be the most beautiful code and it probably will have like a bunch of redundancies and nonsense.

And I mean, this is where like it strikes me that you can then refactor this stuff though.

And like if I've got you know, if I spend say 4 to 8 hours a day building new code and then I can have the thing automated running like refactoring for the next 12 hours, it strikes me that I should end up with good quality code.

But I don't have that process dialled in yet.

Like that's a thing to add, but like it makes logical sense that it can do this.

And when I manually walk through, have it explained, and then refactor code like it can do it.

And there is a process that I can push it through so it can be automated.

Yeah, it just isn't.

Yet, do you know, it's just occurred to me that the two things that we've been like playing around with this week are called hype man and wing man, and we didn't actually intend that at all.

I don't know.

Two men, but I mean I think I will extend the wingman concept to just use yeah, all the different clis yes as well, because it makes sense to just pick them up yeah.

So I use the goose CLI.

I can use the if I can use the core code CLI.

I mean, I do think the I it really is the the goal is to just kind of pull up my phone anywhere, like review an app.

Yeah.

That's that's that.

So do testing stuff, but also just dictate what's going on, get feedback from the agents on what they're struggling with, give them a nudge, like whatever, but come have an idea, pull up my phone, dictate it, bang, done.

And like, I think to do that, I need to like get back to the home machine, as it were.

I think that's the right way to do it.

And then advise it.

So having this like web UI to interact with multiple CLI sessions seems like yeah, seems like Go.

I mean, that makes sense.

Do you know, I think it also allows you to be really selective around what aspects of the code you want to play with the most.

You know, like with Hot man, like I spent a lot of time working through the kind of brand voice Jason and just building that out and actually something I built out in with Core desktop first.

Like I just go back and forth trying to build the personality for for Hot man because, you know, obviously you and I have our own kind of methods and approach.

We've got our own brand voices, but I wanted this thing to be you're.

OK, those I wouldn't.

Yeah, it's just, yeah, exactly.

We're built the opposite of us.

But yeah, I was just like, there's some, there's some like really good personalities on on Twitter that I'll take a little bit of that and a little bit of this and a little bit of this.

And you can like do a bit of research we call desktop and then progressively kind of build out the personality, refine it and then get that into Jason and then pop it back into the code.

That's actually probably where I spent most of my time because I was like, OK, like building this so that it can actually like read through a transcript, do some research and then, you know, post on Twitter all day.

That's one thing.

But it's like, yeah, how's it going to sound?

And how's that going to resonate with people writing it?

Like, it's one thing to just be like, you know, to have an autonomous bot.

That's probably not even the most novel thing anymore, to be honest.

There's so much of the shit.

But I was like, but could we make it entertaining and insightful?

And could it have its own like, unique personality that just becomes yes.

And obviously, like I'm drawing from other people's personalities, but to some degree that is also very human in and of itself.

So, yeah, but that was quite interesting is that I was able to just go, OK, Claude Codes building out the rest of this thing.

Let me just hone in on that one file and just rewrite this in detail.

And that was very manual.

But it was also quite, quite fun, quite interesting because I was just, yeah, spending all of my time there while Claude Code was off doing its bit on the side.

So we're just doing those two things in parallel.

Like I would have, you know, if you were all working together or, you know, we had a team of people around us, so.

Yes, if you had a team of five developers, they would just be they would work, and then you'd go, all right, well, I'll plan the next thing.

It's just the way it works.

It's just kind of nice that the turnover loop is just.

So it's quick, yeah, for this sort of stuff.

And that's what I'm saying.

Like it it's, it could probably be a lot faster, but I almost don't mind because it's so much faster than the alternative.

Yeah, you know, like it's just not comparable to well.

The trouble with going like balls to the wall speed is that you just don't know what's going.

On exactly.

So I mean, that was the, I felt like that was my, I guess sound like learnings from having a few months where I just ran root code like and did everything inside that and I was like, God, I feel like I'm just going mental again.

Like it just doesn't, it's there's too much going on.

Like I can't constrain it in the way that I want it to.

Like no matter what I say in the rules files or anything else, like it doesn't follow it in the same way.

There's something in the way they've done the system prompts that is just like a bias for action, which I want in the execution.

I just don't want it in the planning.

And when I came out of that and when just changed environment completely from the planning, it makes so much, so much better.

Well, and I think because the desktop again is like the agents are made to run on their own, not like, and to ask questions like if they need it, but or for approval.

But really you can just run them full Yolo mode.

And then whereas the desktop app is is only designed to ask to be like a constant backwards and forwards.

So it like one of them is made for, for dialogue and the other one.

And I think that does make a difference.

The like the way they program the system prompts for like a ChatGPT or a code experience is that it is constantly like just moving one step and then waiting for you and it will follow like the process that you set out.

But it's not trying to do the whole thing.

Whereas when you, when you're in the code world, like the goal is to get the code built.

So even if you tell it like, right, we're in planning mode now, it's like, yeah, yeah, yeah, call planning mode, but build code, right?

Yeah.

Because under the hood in like, you know, your prompt is there, but under that is written probably about a million times, right.

Keep going, keep going, just keep at it.

Like fucking like nail it, nail the code, write the code, do the code, code, code.

And it's like, it's like he's like, he's like, he gets him in the morning, gets all that.

And then he comes into the office and you're like, yeah, we're going to do some plan.

He's like, yeah, yeah, yeah.

Code.

He just, it's like, as I said, it's like, you know, it's like a dog, right?

Get your lead out.

And you're right.

We're not going for a walk right now.

It's like, yeah, I can see the lead.

Like, let's there's a lead, right?

Let's go, let's go.

That's so true.

It's just like.

It's so happy, which is, you know, like at the right time, that's exactly what you want.

Yeah.

It's just, you just got to recognize, I think that that is just not the tool.

This is not one tool for these things like this.

Yeah, niches for how you use.

Them it's so funny because I approached it the complete opposite way, but we've ended up in the same place, which is I just want a different environment for my own mental like my own sanity and clarity.

It's like I don't want to be in the same environment building where I am thinking and I'm still at the same screen, but at least if I can segment those environments that I feel like I'm in control of how I'm thinking in one space relative to another, can bifurcate that it's.

But it's you're absolutely right.

Like I think, yeah, especially I've noticed that with code, code as well.

Just wants to go.

Yeah, it's just like this.

Like, yeah, someone's fired the the starters pistol for 100 meter Sprint.

It just wants to just take off immediately and do all of the things.

Oh, that's great.

But yeah, I I find like having a detailed set of design documents upfront is just so much more useful.

Yeah.

I think the other thing is you can't like you, that orchestrator role is critical because you, you can't necessarily like to do that yourself would slow down the whole process like massively.

And it is like you've already got the plans.

It's already done.

It's just you again, you can't let the dog see the plan.

Like that's the that's the dog with the 2 Dicks that he's doing that.

It's like what we call this episode.

So he's doing the implementation and there's the one Dick dog.

The older dog is sat there.

He's the one that later.

It's like the story, you know, the story of the two bulls.

No.

It's like, this is just a, we don't need to say this on the podcast, but we will anyway.

But it's like it's never stopped us.

It's a story.

There's an old ball and a young ball and the old ball.

So the young ball goes, all right, let's just run down there and fuck that cow.

And the old ball says like, no, no, no, let's walk down the hill and fuck all the cows on the way.

It's just like that's the.

That's wisdom.

That's wisdom.

Like there you go, there's a wisdom drop for you.

Sometimes you need to slow down and just do it piece by piece and not get too carried away.

Very good.

This is how we can do the 2 balls.

Approached.

Yeah, that's it.

It's an AI management, Yeah, because.

You need an old bull and a young bull.

You need the young bull.

I don't know if I want to torture that.

I don't know where that analogy is going to go.

Yeah, you need somebody enthusiastic to do the implementation, but you do need like the older ball to just slow it all down and do it because they will just run ahead to the most interesting looking thing, which it's kind of weird that that really human like behaviour just exhibits.

And I feel like a take away for me has just been like, shit, is this how like like I work like this?

What's going on inside my head?

It's just, I'm just like a jumble of like all the things that I see and then like a set of like next actions that I do based on that.

Like this made me think I should be a bit more selective and what I put into my head.

Oh, there's always that that priority.

But yeah, I mean, one of the things I I've also noticed with Claude code is I can quickly, I can quickly tell if it's just going going off on one because it'll have it like marked out in its to do's.

And I can kind of see that coming up.

And I'll be like, I don't know if we need to, we need to do that.

I'm happy to let you do it because you you're probably reading my plan and you you probably know it way better than I do because I've had one like one glance.

That's really the thing.

But I just have this feeling like I'm going to have to walk this back at some point and get you to undo a whole bunch of whole bunch of shit.

But so far it's not been a big problem because as I said, like it's a pretty small, small project.

And I've done that deliberately again, because this is part of the kind of trial and experimentation phase with Claude Code.

It's been fun.

But yeah, I suspect I'm going to have to increase to 100 or $200 plan and just let it go.

I was seeing people like coding multiple projects in parallel as well with Claude Code.

And I'm just like OK, this is this is a bit hectic but this leads to like this.

Almost the problem then, right?

How the hell do you know what's?

Going on, how do you know what's going on?

And I mean the big, I mean, we've talked about this a little bit before, but I feel like we should probably drill into it a little bit, but it's like, how do you know what to build?

You know, because it's just like it's quite easy to get started on every hair brained idea that you've got that pops into your heads.

You can build a plan and start coding it within like got no time at all.

Have something have some working prototype in a couple of hours and you're like, OK, and they'll be there.

You just move on to the next one.

And that's definitely a problem that I've noticed myself having.

And I mean, some of it's good for content, which is fine and some of it's good for learning and that's fine too.

And there but there is this like animus around like what are we commercializing?

What are we putting out into the world?

What's valuable for people?

And yeah, but I don't necessarily feel like I've got a good system for what to work on right now, because I do feel like everything is like dopaminergic.

It's like, even if I'm just learning something, I'm like getting a dopamine hit because I'm learning something and that feedback loop is really short and I can just go out and try a new thing and I'll take the learning from that and build that into another project.

But it does feel very much like it's just exacerbating ADHD.

Yes, yeah.

So I guess if I would look at my my stack, I've got like sets of apps and services that they all, they all get progressively improved over time.

And I'm kind of doing them all in parallel and just flipping between what I'm working on.

But I haven't like pushed anything through into like, all right.

Well, if I was going to just ship an app, yeah, that everybody was going to use, like how do I focus on like that one?

I mean, I think part of it is just like, obviously we commercialize for different means at the moment of just having of a work on, which is then something that you you just need to put the time into and make sure you got the mental space for that.

But yeah, it's, I don't know.

This has probably always been like a bit of an issue of mine, which is that I'm just not the complete a finisher personality type, right?

Like it's always been like that bit where I'm like, alright, I can come up with ideas like all day.

That's what we need to do.

And this is great because I can start all my ideas.

And the the good thing about like the AIS is it does help me finish those ideas like they I do round out the ends.

I think sometimes it's just that there's like the proving of the like the pilot stage where it's just like, can I prove that this works?

Yeah, I can prove that it works.

Right.

Cool.

Now I've got again, an equal or bigger amount of work to finish it off so that I could then launch this in production.

I'm bored.

I'm just going to do.

This a bond to the next one, yeah.

It's just because it can be a bit of a grind to do that.

I mean, I think the, I mean what I also have is that I can have an app that I don't actively work with, which is there as a framework to make it easy to then put these services into quickly launch them.

Yeah, but I didn't finish it off.

I did something else.

So typical, yeah.

But yeah, I think that's like, this doesn't feel like a tooling failure, this feels like a beat failure.

I've just like I enjoy the process of working with these things so much like in that creative state at the start.

I do wonder how some other people will deal with it like so.

The frame of reference here.

There are people that can just finish.

They can just finish.

They start.

Yeah, but it but it's also like.

Maybe they enjoy that bit.

Yeah, maybe.

Maybe they have the like the inverse?

Weird inverse issue, I know.

If you're there, like, please, like, you know, leave a comment like if you like finishing stuff off because you know, we might have a use of it.

I bet, you know, you see this with Jack.

Yeah, like every other week.

There's just like something else that he's pushed production and you're just like, oh, this is like really interesting.

That was a bit chat this week last.

Week A.

Few weeks ago I don't know time is just a weird construct now, but there's just so much being released like every day.

I can't even begin to keep on top of it, but I was like, that's really interesting.

Like this is this looks really cool.

But yeah, the the cycling through of these projects that like he's just got this insane ability to just push the stuff really quickly from from start to finish.

And I don't know how long he's been thinking about it either, to be fair.

But.

I think he's very aware of that as a problem and he is like probably one of the greatest product minds.

Like our generation, it's pretty bit older than us, but you know, similar timeline A.

Little bit, yeah.

You know, like, I think he's like the only person to get two companies into the S&P 500 that he was heading at the time at.

The same time, yeah.

Well, not even at the same time, because Square just came in recently and blocked.

It that was the recent.

Yeah, that.

Like, but he's the only person that's been heading up like as the CEO when Twitter went into SVD and the CEO and Block went in.

It's just, you know, that's crazy.

Pretty impressive.

Like, you know, I think it's his natural personality to be like a corporate leader by any stretch.

But fuck the guy can like come up with ideas and think them through and question.

Them.

So I just sometimes I wonder if that's innate or if there's a system behind it.

Well, I mean, I think again, like he works as I understand it, like he's by coating all these things inside Goose.

Yeah, as he goes.

If you think about like the like what?

Like whilst he is an engineer, like historically, like his like last 20 years must have all been just.

Management.

Dictating and managing ideas through product teams.

So again, like it's probably quite natural to just not look at the code, not see any of the stuff because you just can't anymore, but be quite used to dealing with that and expressing like how to do this or how to do that.

So there could be something that it's just like quite like that's the sort of personality or trait that you need to be in to get this stuff done.

But yeah, it probably also helps that you've got such a large platform that when you launch something, Yeah, that's true.

I feel like I could just launch them and then like.

No one sees it.

Fuck.

Which probably means I should launch more stuff.

Yeah, but OK.

Well, the, the kind of the failure risk is just negligible at that point because the real challenge is distribution, yeah.

And so.

Then interesting you get the distribution because you keep launching.

Yeah.

I mean, this is one of the reasons I'm I'm going to go to the the undisclosed location, yes, for a little while, is that I really want to like hone my getting shit out of the door muscles.

Yeah.

And just launching anyway.

Yeah, well, I mean in a similar way to just releasing the the pod every week irrespective what happens.

I mean, you get all sorts.

Sometimes you get Aussie bird impressions and other times you get hail storms and that's fine.

That's what we get on, on the good stuff.

But yeah, I, I think that there might be some systems thinking here that would come in handy to just, I've been very aware that this has been a problem of mine the last few weeks that isn't necessarily getting any better.

Like I just find I'll come up with an idea and I'm like, Oh yeah, I should just like dive into that.

And then before you know it, you've cycled through a few days of the week.

You're like, oh shit, you know, I'm going to actually commercialize some of these, some of these ideas.

We don't have to.

It's not obvious that it was a better idea than the one you were doing last week.

Exactly.

And like, would I have been better off spent just finishing that off?

I mean, the answer is like, yes, and I know it's yes.

Like this feels like a cancelling session.

We're both like, yeah, I do that as well.

It's like, yeah, it's like group.

So it's like, yeah, this is therapy.

Yeah, I'm sure this is.

But I also get so much like, the reason I know all of this stuff is because I've been following my enthusiasm.

Yeah, right.

Like it's very much like, this is such a new space.

Nobody knows how he, like do any of this stuff.

Like nobody knows like good processes for vibe coding.

I'm I, I, I feel like I've got a more sophisticated setup for vibe coding than yeah, probably most people in the world.

And I feel like that in itself is just good capital to be developing.

Yeah, absolutely.

And I'm, I'm fairly sure that I'll get to a point where I'll be like, all right, I'm going to nail the back end of this process and we'll, we'll have stuff.

So it's always launching.

It's very quick.

It'll go into like some sort of template system and then I'll just pump out ideas at that point.

But I feel like there is a lot to be, I don't feel like I'm wasting time at the moment because I'm just refining this system that is the engine that makes the rest of the stuff almost obsolete, right.

If I'm not great at finishing off the, you know, we were still about like one of the great aspects of AI is you can use it to augment the bits of your character that are bad.

Like if you're bad at creativity, you can lean into it for your creativity piece.

If you're bad at execution, you can lean into it for that.

If you're bad at completing finishing, you can lean into it for that.

I've got to admit, I haven't gone out of my way to solve that that problem.

But I really, I mean, I really should have like this automated kind of like dev OPS constant release sort of thing.

Like at the moment, like I got Beacon to a point where I was, you know, demoing it at the Bush Bash.

And this week I've been like moving it onto a new machine.

So fuck me.

Like like, like this needs to and just tidying up a bunch of stuff.

I'm like, all right, I need to just remind myself, like, of all these different dependencies I've got baked in here that I wasn't probably recording, that, you know, I just sort of rolled with at the time.

And I use that system as well.

So I need to deploy both of those veins.

And yeah, yeah, this feels like a bunch of stuff that is all absolutely automatable.

And it is like in the modern world, I just need to force myself to do that.

But it's also way faster to rectify all of these things now, you know, like you can go the quick and dirty approach to test out an idea quite quickly and you're like, OK, well this isn't like, you know, production friendly.

So but we know we tested it, we've demoed it.

We think there's something here.

Now let's go back and and refine this stuff.

And the speed to go and do that is just yeah, yeah, it's way quicker.

So it's.

Just like, I mean, frankly, it's just less interesting to refine, yeah, of course stuff after the fact and finish it than it is to just come up with a new idea.

And there being no shortage of new ideas to investigate, it's easy to get carried away.

And maybe there's, there's no like external pressure on us to do these things, right?

Like, it's not that, you know, I'm signed up for like a sponsorship arrangement or something where I need to get this out because I'm trying to hit.

I mean, maybe that would help.

But yeah, I, I, I don't know.

I probably usually for me, the way I do it is I just solve it by sticking a milestone on it.

Yeah, exactly.

Usually I sell something for that point in time, like, all right, I need to figure it out by then because I can't let somebody down.

And like, once I have that external pressure of letting somebody else down, it tends to focus me quite well.

I should probably just do that.

I mean, one of the things I've started doing, and it's probably too soon to report back on whether it's working effectively or not, but it's just calendarizing the actual project I want to focus on.

And then that way I'm not like contact switching myself between different things too frequently over the course of like a four hour, 5 hour period.

So I can just do this one thing.

And obviously, like in our defence, we've also got like a bunch of other work that we've been doing this last month or so and that's obviously been taking up a lot of the, a lot of the oxygen.

So we've focused on that and that's coming to an end.

And so, you know, we will have more time and space and energy, I suspect.

So to, you know, get a few more of these like smaller coding projects finished and up and running live.

And and so that will be good because I also think that will give us a bit of momentum as well.

When you actually see something out in the wild, I imagine that will be quite, quite energizing as well.

So those things are coming.

And yeah, I but I, but I would be curious though, if that at least helps.

Part of the challenge, of course, is, you know, sticking to the calendarization and not just going off the reservation every time I get a brand new idea, which is the other thing.

So how do you caption new ideas when they come up?

I will like like generally have something half baked again, just go with, go into it with Claude and just basically go and prompt Claude into being like a bit of a idea co-owner with me, but just to help me flesh it out, ask questions.

That you always flesh out the ideas as they come up.

I wouldn't say always.

Sometimes I'll just sit on them for a bit, like just something that pops into my head because it's still pretty half baked.

This was actually one of the things I found interesting because I kept hitting limits.

It will just basically lock you out for two or two or three hours at a time.

And that's been handy because then I'm like, OK, I've actually just got to close this thing down and go do something something different.

And so it's like I've been like a forced mental break.

Because I mean, we've talked about this before, but like one of the things I'm, I'm really cognizant of is not abdicating thought to language models.

Like I really want to own the thinking piece because that worries me a lot.

Actually, just at a macro level for how people use AI.

It's just like, you know, how easy is it to just basically abdicate all of your your thinking thinking work to the language model, but.

I feel like it's easier to abdicate the action work of like the knowledge work, you know, like coding is kind of like the typing with your fingers is, is, I feel like is going away.

But I don't feel like that's that bad.

Yeah.

Because like that was always like the means to the end.

The point was to create the product and the thing that we needed, not to necessarily just craft code.

Yes, so.

So like, I do feel like there's that's OK.

And then the bit that you need to think about is the like, what am I building and why am I building it?

Where we start?

Yeah.

Which is the thing.

If the cost of doing something goes to 0, then the only real constraint you have is like what you should do.

Well, yeah, exactly comes back to that question.

All sorts of stuff.

But like you know, I also think this is the main main issue with code.

I mean, like a lot of the stuff we can do is always achievable using software.

It was just a case of like, what was it?

Was it worth it?

And we do a lot of things that are just bad ideas that people would spend a long time on or we would do.

There's good ideas that nobody really gets around to building because they don't know if there's enough or the wrong person has the idea that they can't implement it.

It's just.

I feel like the, the, the bit that makes it the wrong idea is the opportunity cost.

Because traditionally you would have spent six months a year going down that track before again.

And maybe you're like on the fence about the idea or maybe you fully have bought into it, but it turns out to be a bad idea.

But like the issue is you've just wasted, well, not necessarily wasted, but you've spent a year or six months and it's like, what else could I have done with that time?

And that's the hard bit.

But we just like narrowed that window down to like a few hours.

Yeah, that's and that's a great thing.

If you can just like capture the idea, get something put together to test it.

Like I, I mean, I as we think about it, I'm just like, yeah, I really do.

I need to like just nail in this light release.

Get the one thing.

And that's the thing that's going to make it well, it's not like just focus on one thing, but just like nail in like, OK, how do I get these things out to test?

Because what I'm I'm like testing them myself.

Yeah, yeah.

I mean, like, so it becomes a I then become my own bottleneck of having ideas, testing them and go, no, I like this one.

Yeah, great.

Yeah.

But like, I'm not like, I mean, it is it is meaningful if I helped myself do stuff like that is good.

And I think This is why like the coding space of agents is so much more advanced than everywhere else than us.

It's almost like, why if you want to be interested in AI, you kind of have to get into coding just because all the people that are really into AI came from like a code space.

And then that's the place where they've really pushed the needle on like what agents can do and how.

And you discover all the different ways that they work.

And then you can translate it into something else.

But it's, it's people just solving their own problems, which I, which I do think is the answer to this question is like, well, what do you build?

It's like, well, you, you know, you, it's hard to like know other people's problems in any detail, but you absolutely know all yours.

So it's like massive acceleration in like software engineering coding because everybody went and solved their own problem for I can't write code quick enough.

I'll just do this like, you know, all the slow code stuff that we've been developing is really is like kind of like, all right, well, you can write this code, but I can't control it in like a meaningful way.

So I need solutions for that.

And then, you know, we've got to that point where we've got a kind of setup that works quite nicely.

I feel like probably just need to like step back and go, you know what my current problem and bottleneck is that I can create stuff quickly, but I'm not deploying it quickly.

So I need to look at my process and say like, how do I do the back end of this process and how do I get this?

Just dial in so that I can get this out.

So maybe I need to do that before I solve being able to walk around and come up with new ideas because.

Because then I just end up with more new ideas that are deployed.

So, yeah, yeah, that's interesting.

The backlog just gets bigger.

Yeah, that is right.

That's valuable.

Yeah, I, I hadn't like, I guess, you know, we've been busy the last couple of months as well and short weeks have been away.

I've had like more projects going on, but yeah, like, yeah, like this is maybe it's useful to have a podcast.

So we have nothing to discuss because I, I have personal realizations and we just go like, you know what, that's the thing.

I need the, the release part of my, my cycle.

I'm sure we're not the only people thinking about this stuff as well.

Like one of the other things kind of loosely associated with this that I've been thinking about is like how much better does the does the tool need to be or does the environment need to be or does the experience need to be?

I kind of feel like like where I'm at with Claude code, I'm just using Claude Desktop.

I could get a lot of stuff done and I'm like, do I need more?

You know, I see new releases and like new ways of working and people running, you know, like writing code across multiple projects with Claude code.

And I'm just like tell you that I need that.

I feel like I can get enough done.

That's still way faster than pretty much 98% of the population.

I mean, even my idea, I'd like to be able to control it from my phone so that I can.

Peek in, step away from.

It like it's like on moving from two different computers so on my laptop like I do a lot more approval of what's going on because I care about that machine where it's on the mini.

So yeah, it is just full yellow mode.

So I'm like, I don't have other stuff on here.

This is just like, this is your playground Go.

Yeah.

And it does just execute and I don't have to keep like moving along.

So now I'm like, you know what, Maybe it's not such an issue that I need to do that.

And maybe I should just, you know, reframe this and say like, yeah, it's like if I have to, when I'm doing the planning, maybe it's correct that I should have a laptop for that.

And I sit down and I, you know, make myself comfortable and I have a nice cup of tea.

Yeah.

Get myself in the right mind.

And we, we meaningfully create a plan of a feature.

And then I can go and go for a walk while something else happens.

And at that point I'm just walking.

I'm like, because the way I would then like capture ideas would be like, I, you know, I'll be going for a walk around.

I'll just fire up.

Like if something comes into my head, I'll just go into like ChatGPT and just dictate straight into it.

Often on like deep research mode, I'll, I'll set up like, OK, this is what I want to do.

I'd really love like a project where just give me a tutorial that tells me how to do this.

So I want to try out this new thing, Give me a tutorial that implements it and takes me from start to finish, and I'll have it.

I've got tons of these, right?

Just create them and some of them I'll come back to, some of them I won't, but it gives me a place where I can come in and get all right.

I want to see this in action.

This is like my piloting process.

So I get that tutorial and then I'll just have root code implement it and follow it and then I'll look at it and then I'll often just throw them away afterwards.

But it's a good way of like capturing stuff into a form then works like that.

And I could probably like automate that.

So it starts off while I'm not a while I'm still out and about, but it's not an issue.

I can just file them off when I'm ready.

But maybe, maybe, maybe it's correct.

And it's like, that's the way, the right way to do it is say, all right, I'm going to dedicate a few hours every day to getting in the right mindset to writing the new features and the plans that need to come in.

I'm going to be more or less hands off with the code.

I, if I then, you know, maybe I could just dial in the deployment thing so I can test from my phone or something like that.

And I'm out and back and I can spend the rest of the time just gloriously desk free.

Yeah, exactly.

Yeah.

And like one of the other things is when, when I'm certainly talking about ideation with language model, it's not necessarily like a big macro idea either.

It could just be we're building something and there's a feature that you just want to flush out in more detail.

And it's like, would this be valuable?

Are we just having something for the sake of it?

I find out myself doing that like a lot.

We just ask questions and the other thing, I know I've mentioned this in the past, but like in my prompt, I will always be like, OK, just let's tamp down your agreeableness because, and you know, I will just put like a percentage slider in there and I find you'll get pretty different outputs.

Like sometimes I've like dropped it down and just said like just go to like 40% agreeableness.

And I, it's some of the questions that will give back to me.

Like at one point I was just like, are you just being Dick?

Like this is, this is just not valuable at all.

And you know, because it was just trying to be a Dick basically, rather than trying to be insightful or valuable.

And I was like, alright, well, just let me like dial up that number a little bit and see if that changes anything.

And to be honest, I kind of did.

It was quite entertaining, but really, really valuable to just like step through steps, through ideas and force you to think about if something is actually valuable or not valuable.

And again, like having code, code write, write this stuff and then do a review in a different environment just gives you that like that clarity to just, yeah, freely assess it.

Because otherwise I could be like, you know, in code code and be like, do we need like cost comparisons?

And it would just be like, you're absolutely right.

We don't need it.

It's like, well, hang on a minute.

Like I wasn't objecting to it.

I was just asking a question.

So yeah, sometimes I find I find you need to like just step out of that environment and think through this stuff.

And yeah, because the other thing is what I've what I've done in high man as well is the different sub agents we use different models.

So we can use DeepSeek for like a heap of it for free essentially.

And then for some of the like reasoning work upfront that uses Sonnet for and that seems like it's a nice balance.

But going through stepping through this over like the course of like half a dozen different iterations, it was giving me cost estimates per transcript from going like this is going to probably cost like 20 to $30.

And now it's down to like $0.50 per transcript from because we've been able to use like DeepSeek as opposed to having everything done by by Sonnet 4.

And yeah, so that was interesting, like going through the process which stemmed from, do we need like cost comparisons?

Yeah, which I wouldn't have.

What I didn't think about at the time, it was purely coming up from a question of do I need this like reporting in this in this product.

No, I don't.

I don't need this.

This is not adding anything.

I just want like a a quick cost line that tells me how much we've spent.

That's it.

Yeah.

But it led to so much more, which was quite valuable because then I was able to go and like, you know, reorient all of the open router config in the in the in the product as well, which was quite useful.

So does that give you is it like in the pipeline network or just go like, all right, this, this run and this conversation cost you like 2 dollars, $5 or whatever?

Depending on yeah, but it's not a conversation.

It's basically like 1.

So there's like an orchestrator.

It's like a set of runs, like, yeah, exactly the different agents that happen, yeah.

Yeah, well, there's four basically.

You've basically got one that is the orchestrator, another one that's just ingesting that transcript and chunking it up into themes.

And then you've got another one that's doing a bit of research, which is basically just like a web search against those themes.

And then another one's writing out tweets.

And then the final one is actually just a publishing agent.

And that's the bit at the moment where I'm like, oh, what's the best approach to do this?

Because currently I've just.

It's got like the tidefully API.

It's a bit annoying but it might work.

I don't.

Know I think the best thing to do is to just just use my nosta stack and use that.

I only publish the nosta and just abandoned all social media could know to be allowed to use it in the next 6 months anyway.

Well, that's that's what it would bother God.

It will just be bots in six months.

I mean, it already is like the interesting thing is that I will post stuff to Noster and Twitter.

I think I might have like marginally more people on Twitter and like because of some work we were doing last year, I got myself a blue cheque.

But I don't think I'll ever, I think I might even just delete my Twitter account.

I don't know.

But like when I actually post stuff, the thing that gets like real human interaction is on Nosta and like you, you see people react to the face.

It's you have to time it because there's no algo that keeps exactly stuff rethinking.

But it's almost perfect for just a bot that just posts like every hour.

You can post the same thing like once an hour and then people might see it when they come online.

But like you actually do get proper interactions with real people, like people have discussions and the same way that I remember they used to like in 2006 on Twitter, like it's just, and I just don't think you get the reach that I think you get a lot of fake reach on these platforms because there's advertising dollars behind them.

So they're they're incentivized to lie to you about how many people have viewed whatever or like what likes you're getting.

Like bots are good for Twitter.

Like it's, it's good for them.

It looks like there's more people and they can sell ads on it.

Whereas like, nobody really profits from that punching box.

So.

People just ignore them.

I mean, they definitely are like the same.

I feel like there's five people that every time I I load it up, it says these people have followed you and I'm like, oh, just fuck off.

Like I just don't need to see this.

Yeah, it's notification every time.

But like, yeah, like that's where I get all my meaningful interaction, you know?

And then people send you money.

Yes, You're like, all right.

So.

Maybe we just.

Post the content there and actually get people sending.

I mean like if anyone is listening, if you don't use nosty you probably, I feel like you probably do because I'm not sure we really push this enough to do a bit.

But get on there.

Just go download primal or something.

Myself included, to be honest.

You really?

Yeah, you really should.

It'll be a nosty game.

It's not, It's not.

It's so nice when somebody just sends you money.

Good for a nice tweet, even if it's like negligible amounts and.

Well, I think get into this.

Habit of going yeah, I could like that or I could send that person 21 sets.

Yeah, I'm just going to send.

I'm just going to send the money for that nice tweet that I like.

It does feel like a nicer environment as well where people are still sharing their ideas.

And yeah, well, it's because it's full of people.

Yeah.

Like that's the difference.

Like it's like, yeah, remember when Facebook became full of, like, brands?

Yeah.

And it's like, that's when I left it like, I don't know, 16 years ago, 10.

No, no, maybe 10, 10 years ago or something like that.

And it's like this is terrible.

Like this is just like I don't care who likes Mercedes-Benz, like yeah, sure.

They don't even feel was not even a thing.

Like, just like in the shit up.

Prior to that point you would only see content from the people you followed.

Yeah, it was your friend.

Literally.

Also 2006.

Yeah, it was about the time I moved to London where it like, really hit stride and you're allowed in even if you didn't have the right university e-mail address.

And all of a sudden I realized there's like another, like, you know, 20 people that were living in London that I knew from uni that we started, you know, and then you start meeting up, going out to nightclubs and stuff, and it's a lot of fun.

I don't think that point of life would have been as much fun without Facebook.

But then very quickly it became rubbish because I got older and older went from being like a good social coordination tool.

I mean these days it is good for groups like annoyingly like all the groups on there, but.

Well, I find people still run all of their like events and stuff through it, which is really irritating because I will miss it and then someone will text me and be like, have you seen this?

I'm like, no, thanks for telling me.

Yeah, exactly.

It's.

Frustrating because it's the sort of stuff that could be anywhere.

But I think that became such a focus.

Like that was the place that everybody got an account exactly early on that it's like the shelling point for events and community groups.

And you know, like weirdly, like all the sort of hippie homeschool stuff is all on Facebook.

Like this is such this is just not the platform for you guys.

What are you doing?

You should be on Noster like, you know, getting everybody.

So we need to double down on Noster.

Well, the good thing about nostrils, you only have to make one more change.

Like it's not like you were in Facebook and then you went to Twitter and you went to LinkedIn and you went to this and that and they're all separate.

It's like once you get to Nosta, you've made one change and then you can have the Nosta version of LinkedIn, Facebook, Instagram, everything and it will just be the one account forever.

So it's like the last social media migration that you need and then you take all your people with you wherever you go.

So it's, it's a nice lower lift.

You just got to find different ways for people.

I mean, that's like Beacon gives you a, gives you a way in, you get a Nostra account with that that you can take with you.

And then, you know, you could just be using it as a Bitcoin wallet via WhatsApp.

And then all of a sudden one day you log in via a machine like a web browser and you're like, what the fuck?

There's there's like all these events and stuff going on in here and you get it by surprise.

Yeah.

But like, I do think it's pretty the place to focus the energy on these things.

I agree.

OK, we'll double down on it.

We'll need a hype man for for Nostar.

Yeah.

That's going to be, yeah, the next idea.

And the general view there as well is that there just needs to be more content and specifically like more content that's not about noster and Bitcoin is that the two areas that like it's really big for people talking about how great it is.

Like, you know, that's that's got a bit reached, but and then people that love Bitcoin are all there as well.

So like there's like a which you need like in any sort of new network, you need like a founding group that are going to go over there and take that conversation.

But like there's not like there is some AI stuff on there, but there's not tons.

Like whenever I manually go and post the the interviews or like experiments we're doing, but it goes really well.

Like it's, I think it's it's good content for that space.

So if we had a better, I mean, I was thinking that you've been just building out a quick open source like publisher where I can go.

OK, well here's I'll write the tweets into this, but please the posts into this but Nostra and then just post them like every few hours because I yeah, I know it sounds spammy, but yeah, because there's not this algorithm that keeps floating everything back up like unless people see it there.

Yeah, I guess lots to.

You know, half the people I'm connected with are all in the States.

And yeah, yeah, that's different.

Do I really want to stay up late at night in order to get better social media?

Well, probably not.

But like some of this stuff I could put out and then.

Yeah, just content recycling makes a lot of sense.

And but what's interesting about like the two examples you gave is that they're overtly positive.

Like the reasons for using nostril, all the types of content that you see do well on nostril, they have like a optimistic or positive undertone.

Whereas on Twitter it's like it has to be rage bait in order to get impressions.

Everything else kind of just like, you know, you could you could buy something incredibly valuable.

Strange slide under the radar or you could post something that is designed to get as much visceral hate as possible and it will it will just fly.

And that's the bit like I just don't want to constantly be in an environment that is just bombarding me with the same negative shit endlessly.

So I get it.

Let's do something about it.

So we may as well like sit in Noster and OK, that's that's my goal for this week is just get more more active on on Noster.

I get hot man up and.

I think I'm going to have to work on finishing, completing, and finishing.

Yes, that too.

Well, that, that feels like that's just going to be like a, a constant, a constant goal.

Deploy and ship.

Yeah.

But I think the like with me, when I deal with these problems, I feel like it's, it's just acknowledging that I know I'm not going to naturally do this.

Like my my tendency and the where the energy is, is always going to pull me to.

All right, let's let's start something new.

Let's follow this, Fred.

Let's see how this works out.

Let's experiment with this.

And I mean, even just, you know, shipping can be as simple as that with the pipeline of stuff that I was building.

I really like that as a tool.

And I I think there's places that we could take that.

But like even the first versions of that were as simple, like the shipping was as simple as.

Yeah, just post like creating a wiki on the GitHub and just being like, all right here, here are examples of these conversations actually being run.

Like look at the difference in like how the bias comes out.

And these you run like 2 Sonic models for 20 runs.

Look at how that their bias and their system prompt comes out front and center and you can see it in the conversation and like where it goes and how they steer it because they they take it on a path that is just not included in the source material.

Like that has come from somewhere and it's been reinforced in every run.

So why?

And you can discover stuff like that or you take the panel format and you say like, oh, wow, this is like actually discussing the thing from several different angles and coming up with interesting observations.

Or like just the dialogue coming up with new knowledge where you take something that's like, if it's not contentious in the system prompts and you let it run and it actually like you go, shit, that's a great idea I think of.

But like just, I think just getting this stuff out, like I think we've talked about it a bunch, like the best content and marketing for ourselves is probably just talking more about the stuff that we're doing.

But we just, I don't know, it just doesn't sit with me as well as I just assume nobody cares.

Well, I.

Don't know.

I guess also an afterthought that finds like, I'm not really thinking about it in the moment.

I'll think about it later and be like, oh, that would have made really good content, I should do something with it and I'll make a note to do it, but I won't actually create the content.

Yeah.

And push it out.

So yeah, there is a there.

Again, this was probably just there's a system to this, or maybe it's just a case of doing it in the moment like instead of making a note.

I feel like saying like I just need to do that is a bit like saying like I'll just have great willpower of a diet and something.

It's just like, it's just, you've got to sometimes you just got to recognize things and go, all right, this is probably a flaw.

I I should come up with a habit system, something that just removes this so I don't have to try hard.

Like definitely, like there's been weeks where I'll I'll, you know, I'll have like a few weeks where I'll just like do a video face to a camera like every couple of days and it goes well.

And then I go, yeah, OK, I can do this and then I go, but I don't want it just kind of, there's just no energy there that drives me to it.

So I could try really hard and do it, but it doesn't, it doesn't work.

It's just not sustainable in the right way because as soon as I'm busy doing something else, I don't do that.

Like it's just yeah, yeah.

It's funny, as we were talking about some of the social media nuances, I immediately started thinking about LinkedIn.

LinkedIn is like an altogether interesting place.

Like, I just don't think, you know, like, I think some of that content was super valuable.

It's like if someone could basically like come back from the future and tell you how the next like few years are going to play out, but you knew they'd come back from the future, you'd be like, oh, I'm going to pay attention to this.

But The thing is like most of these people on LinkedIn, like he could post this like incredible alpha, but they just don't, it doesn't, it doesn't register at all.

It's like if I post that I've just started a new job and other stuff, I'll get like 100 likes and people will be like amazing, congratulations, you know, all that kind of crap.

And then you've got like 10,000 impressions.

But you could literally be going like, this is how we're using this tool and we just created all this leverage by doing it.

And here's how you can do it.

And yeah, just nothing.

I mean, it's, I was going to say, I feel like we maybe have a little bit of a love hate thing with LinkedIn, but I'm not sure there's that much love.

No, but it's like, I mean, it's certainly it does work.

And I think there's an interesting aspect of LinkedIn is that it's the it's the most geographically segmented of the social networks.

Like in a lot of other social networks, the people I know aren't here in Perth and within the majority of the people here.

That's true in Perth, yeah.

And then, you know, I do want to work with people in Perth generally because it's just it makes it, it's more fun to be in person with people.

Yeah.

So that that it's like that's the only place I'm like, I mean, I guess Instagram is very bit seems to be a bit like this as well, but.

I feel like I see a lot of content from, like a lot of people I don't know.

Yeah, it's I don't.

Use Instagram so I don't know what.

It's like, no, I'm speaking about LinkedIn.

I'm thinking about how to work this, but I find LinkedIn is predominantly people I don't know.

Yeah, and every so often I'll get a post surface from someone.

I do know what the post will be from like 2 weeks ago.

I'm like, I would have been handy to have seen this two weeks ago for for that benefit.

Right.

I think it works for like if you've got like something that you need to, if you're doing like consulting work or something like that, where it's likely that you being physically proximate in this space and they need to find like who is the person that lives in Perth that does this?

Like it's really good for that to be that person and you can just talk on it.

But like I just, I find the whole the whole site to be just like a tad acky.

It's just full of people that everybody's kind of selling like a fake version of their career in in a way, like the stuff that does well, it's not like it's not that it's useful information.

It's something that you can say that somebody else can add to to make themselves look good.

On a subject So basically like a corporate environment.

Yeah, I.

Just like, yeah, it's like when you know, you have like somebody has a good idea in a meeting.

And then we.

Can go around the room allowing everybody else to agree with it and say, Oh yes, I also agree with that.

And and we add like an inane thing that just never needed to be said.

Yeah.

But it's, it feels it's got like that vibe to the place.

And I just, I can't, like, I mean, in terms of life stuff I've got no energy for like, I mean clearly, like I just barely log into it.

What's really fascinating is this like incredible bifurcation of views around AI that you see on Twitter versus LinkedIn.

Like, I feel like all the discussion about AI on LinkedIn is just privacy and governance and how to regulate AI.

And then everything on Twitter is just how people are using this to create insane leverage.

And I'm like, when I see a lot of that content on Twitter, I quite like Twitter.

And but the algorithm will keep feeding me a bunch of other stuff every so often that it's just designed for rage.

And that is also my own viewing habits of feeding that too.

Like I'm, I'm responsible there too.

Like it's easy to go down these tracks sometimes.

That's where I feel like I need to just pull a pin on Twitter for a while.

Like it doesn't.

I don't believe it serves me.

But I do love it anymore.

Some of the like use case, like I've been seeing a lot of really cool like vibe coding use cases for even for clawed code like that.

I want to go and try and I'm just taking note like bookmarking a shit ton of tweets and going oh, I need to try this and just see how it works and we just.

Have an agent review Twitter on our behalf.

Sort of things just like go and look through my followers and just like just pick out.

To be honest, you could pick them.

Like pick out interesting stuff about AI and let me know about that.

Like scrape it and send it back to me.

I find it tends to be the same followers as.

Long as I don't have to log in, yeah.

That would be really useful, I think.

But yeah, that whereas on like LinkedIn, I never say this.

I just say like, we need to regulate this and government, government needs to do that.

And I'm just like, what?

How, how like these two environments creating like polar opposite perspectives on this stuff.

But yeah, we could sit and hate on LinkedIn all day.

It's probably not.

Probably.

Not I don't know if that's.

Probably not valuable to anyone, but.

I imagine everybody already hates LinkedIn in their own way.

I suspect if you're listening to this anyway at this point and I've listened to all of our episodes, probably predisposed to a lot of the same stuff we are.

But so lesson #1 here is going post more on stuff.

Yeah, I mean, Jeff Booth's been saying this for a while now.

It's just like you've got to go and spend all your time and energy in the system that you want to win.

Like just don't, don't give it to all the other crap.

I think he's right.

I must say, I do love the idea of just being able to send people stats for, you know, when they just post something that's really valuable.

Yeah, I need to get better at finding new people to because that's follow as well.

Like you really do need to follow people to get an active feed inside.

There's a nice incentive mechanism there to actually go and be thoughtful, produce really interesting content, really valuable content for.

People and you'll definitely say stuff and like nobody will just nobody will see it.

Everybody happens to be asleep at the time.

Yeah, of course.

Definitely a lot of the like the timing and the stuff like a lot of the activity is in AUS Time zone makes a huge difference.

Yeah, but just got to probably need just more Aussies.

Well that's the other thing.

It's like what gets more Australians onto to nostril and place.

To go without signing up to have their face scanned every 20 minutes.

Yeah.

So in December, basically pretty big uptick in, I don't know.

I just, I lose faith with society sometimes.

Yeah.

I don't know why we're sleepwalking into just a mass surveillance system and accepting that it's a really bad idea.

Like, I don't know, having met, like enough people that live in mass government surveillance systems, like, they're bad.

Like, like all the worst atrocities in the world ever.

Yeah, were committed by governments like they're not, you know, there wasn't like there wasn't an individual behind the Holocaust.

There was a government behind the Holocaust.

There wasn't an individual behind, you know, the starving of everybody in China or everybody in like the holiday more in Ukraine.

Like it's just it was a government behind all of those things.

Like it was a government behind Chernobyl is a government.

Like it's just it's it's the worst thing is to aggregate so much power to so few people with so little questioning.

Yeah, it's just insane.

Well, the other the other side of that here in Australia is the growing dependency on government.

Well, it's only employed by.

Government, 80% of the job creation was government related in the last few years and it's just insane.

They're insane numbers.

Yeah, you continue to have this like dwindling private sector, which has probably been in recession, to be honest.

If we just took the private sector in isolation, it's probably been in recession for a couple of years and then the rest of the economy is essentially propped up by what, mass migration and property prices?

Pretty much it.

Yeah, property prices and government.

Government spending and.

Spending, Yeah.

And just yeah.

God, what a grim way to yeah.

Come to a point.

It's not not amazing.

Like, I mean, the nice thing about Australia is that it's, it's absolutely, it's still like a beautiful, sunny, high trust environment.

And I'm a massive fan of those things.

Yeah, but I, I think like if we try to find a like a positive spin on, on all of this stuff, it's like there is this like incredible opportunity with AI to just go and create like infinite leverage, you know, like it's, it's just, I don't know, I find it really energizing.

Whereas I still think there are definitely days where I'm like, Oh yeah, we're all screwed here.

Like it's just going to be no ability to earn income in a few years time.

But then I'm like, I can just go and build a bunch of shit right now.

Let's do it.

You know, it's very energizing.

We just got to finish.

Let's ship some projects.

Yeah, that's some.

That's the objectives.

We report back next next week's POD.

All right, let's make next week's pod.

Next week, what's going to be a report?

On what we've been up to.

What we've gotten in after we've.

Just.

Introduction.

To all these things, all right.

Maybe we're not back.

A bit of a hit list, like I'm going to do like a schedule.

Ethanol sticks.

I've been one.

Yeah, that's cool.

And I have a lot of the stuff kind of part built anyway.

I could probably get it up and running pretty quick.

That would be, that would be handy.

Yeah, just to be able to recycle your content as well.

Figure out like what?

Like it'd be cool if you could go and pull in impressions and just go, oh, no one saw this.

Like, let me just like repost that.

I don't know if that's remotely possible with no stuff, but yeah, something like that would be cool.

I I just I probably just post the same thing.

Or maybe it's the the opposite.

Maybe it's the maybe it's the stuff that does really well.

You want to go and post that again?

Well, then you can repost it as well, like it's, I mean, you, you, the whole thing's just like a wide open protocol.

So I can do whatever you want with it.

Yeah, cool.

That's the beauty.

But you don't have to ask anyone's permission.

You can just go and build whatever you want.

So it'll probably be useful just to have more tooling like that in general.

Yeah.

And it's the only social media left.

Let's whip it down then.

Let's get ahead of the curve finally.

That makes sense, it's easier.

Than going and joining fucking TikTok.

I'm not doing that.

Let's get ahead of the Curve could be a nice little podcast title.

I I think for this one, it should be deploy projects like or ship something.

Just finish.

Just finish something.

Just ship it.

You can just finish things.

Yeah, you can't.

We'll just add this to the the merch design queue as well, I think.

All right.

Nice.

That's a good color.

That's the good stuff.