#5 Browser Wars, Cursor 1.0, Software Engineering in the Age of AI w/special guest Chris Addams

You might have noticed that the man before me is not Iranian. It's not Kabaza. And no, I don't think either of you have a beard. This is Chris. Say hello, Chris. Hello, Chris. Hello, Chris. A classic classic. Uh, so I' I've invited Chris on the show. If those of you that don't know or brand new to it or whatever. So Chris and I used to have a podcast like years ago and would you believe I think we were running for about two years you know we were we did I think 74 episodes and we did excluding the news ones because we did do some news shows and we did do news ones and yeah thanks for the reminder we did do news ones every Sunday we would kind of just do tech news wasn't it?

It was broad tech news and whatever and uh so we're very familiar with this. Um what microphone are you using right now? Uh, I think I'm using my old Blue Yeti. Does it sound all right? Say, yeah, sounds all right. It sounds maybe a little bit echoey, but like Well, well, I mean, yeah, it is a little echoey in here because the cat that did just walk across my screen a minute ago is a stray. So, we have plastic down everywhere and no soft furnishings cuz he's a stray. So, it's like a kill room in here at the moment. Nah, it's for you, isn't it? Because you're always falling over. That's what it is. Yeah, I wetting myself.

That's the problem. Yes, it's for him. That's the issue. So, um, it is a little echoey in here than normal. It's all good. But, uh, yeah, so we had a podcast for a few years talking about general tech and doing news and stuff like that. So, you have the techn, you have the equipment, you have the, you know, the confidence to just appear on a podcast last minute. And I wanted to just to get people on the show every so often just to kind of share in the news and chat about AI. And and funny enough, I think you were the person that probably were the was the first t first um time I heard the words GPT and I can't remember what episode it was, but I think you I think we were talking about something.

Maybe it was the AI girlfriend that I had and I think it was used to use GPT 2, I think. Anyway, 3 I think the the start was I think. Yeah, but I Yeah, I just have this thing in my head about GPT4 uh two, sorry, two, but um anyway, so very familiar um with tech and you've done software engineering and all the rest of it. So, just felt quite natural to bring you on the show and and chat about the news. I wanted to say something as well. I've just totally forgotten. Anyway, I remember diving into the uh into the AI as soon as it was open and as soon as ChatGpt was a thing and you had to sign up for for a beta, you know, I was on that beta list, you know, I think we talked about being on the beta list and applying for it.

Um yeah, so I think we we're both being early adopters of uh of the AI thing, the AI world. And I remember what I was going to say. Uh we more specifically you had uh what was the second name? Blake Le Moine. Le Moine from from Google who actually lost his marbles and and announced that Google Gemini was well this before it was called Gemini, wasn't it? Google was uh sentient. We were probably one of the as far as I'm aware what what he was claiming was sentient. This was way before Chat GPT came out. So what he was uh what he was claiming was a sentient AI. I think he's pretty much I think he said since is nothing what we've got now is nothing like what he actually experienced internally.

Interesting. So I think there's still supposedly some other stuff that sits behind the sc behind behind the screens I suppose is probably more better than better than behind the scenes that something that's behind the scenes that isn't um isn't quite out yet. But then again, I think the the thing that's quite interesting about it is you can have a conversation with it and and think that it's real. You know, certainly for um for for lay people, it'd be very easy to have a conversation and be convinced and coerced and whatever it is or you know, blackmailed as we've seen in the news this week um by an AI. You could be quite easily convinced I think.

Yeah. Yeah. And you know, I'm so so in tech and I'm, you know, yeah, I've I've had conversations and, you know, a lot of people use touch as a therapist even and I I've been like I can I can totally see why people think this is sentient. It's so convincing in that respect. So, and uh yeah, so all that to say that the the podcast did did fairly well. We had a good two-year run. We did. Um and that was one of the episodes that blew up. I don't know how you managed to get him on the show, but we were one of the first people to actually interview him and it helped. Um, yeah, he interviewed he interviewed on that tech show at the same time as he was interviewing with like Bloomberg and NBC.

So, I think we I think we did get quite a good scoop there. Yeah. Yeah, for sure. Um, anyway, let's let's dive into the news and see where we go because uh yeah, for those who just joined, Kabaza has had to shoot off for um uh family stuff. So uh if he's watching in the in the in the arm of his chair somewhere then hello KBA say hello in the chat if you are around but uh this is Chris joined us on the show this week and we're just going to get into the mostly we talk about try and talk about web stuff like mostly around design digital and web but we'll dip into other things and the first kind of um interesting thing that happened this week and unfortunately we haven't got much to talk about is actually the DIA well more specifically can I do this?

There we go. Um the DIA browser which is an AI based or AI focused browser and um the owner Josh or CEO went on to the waveform podcast and and and spoke about it and kind of give us a lowown and this is something I've really tried to wrap my head around that overall like I found out the do you what browser do you use Chris? Do you use Arc? What? What? How are you Devon these days? I'm old school. I'm using Chrome. Oh goodness me. Oh goodness. I mean Chrome to be fair. I do do a lot of development, right? And most of the things that you're trying to support are still Chrome and Firefox and uh and Edge.

Yeah. Oh, fire. I mean, I don't I mean I use Firefox, but I don't know what's uh what your your um clients are using, but yeah, I guess uh I mean Arc. Have you tried Arc? Uh, I haven't, I'm afraid. Okay, so Arc, I mean, you can kind of see it right in front of you. You see how this window is kind of full screen right now. And actually, if I just do this, you'll see it pop open on the like I think it gives K like it it it's more of a full screen experience for exploring the web. And because you've got this sidebar, which I can pop open permanently there, um, I think it's more suitable to websites to be like this rather than having a kind of tab bar at the top that's pushing everything down and shrinking everything.

So I think this was like a an attractive feature to people who always on the web and Arc was the browser that kind of introduced this and it's chromium based. Um and it got really really big through this uh interview on the waveform podcast. Let's uh Oh god, there's audio, right? Josh was on the podcast here talking about cuz they basically abandoned Arc after maybe I don't know three three years. And this is the CEO and he's building an AI first browser called DIA. And I kind of like the idea because basically it encapsulates everything all over like everything that you do rather than being like a window or what as he says here an app or a button.

It's an entirely new environment. And given that you're still on Chrome, I wondered if you could wrap your head around this idea that AI is more of a an encompassing thing than just a chatbot or like a like a I don't know, as he says, like a button or whatever. Um, well, it depends how you I mean, how are you using AI when it's within the browser like this? So, what is the actually we can't download it because it's they're really secretive like this. So, I I haven't actually got access to it yet, but like um we'll go into opera actually, but maybe there's something in here that it's it's supposed to it's kind of his grand vision's kind of um uh context like personal context of everything and it's supposed to understand the web pages that you're on.

It's supposed to, you know, understand and start learning the way you write because you live in the browser. You're filling out forms. You're writing emails. It's learning as it goes. It's like your own personal like Google algorithm, which on the one hand is quite nerve-wracking to be honest, but on the other hand, it's um if they can if they can reassure the privacy fiends out there, then um it could be a really interesting experience, but I just I I I'm struggling to think about I mean, you can see this little window here. I've got some auto browsing. That's weird, isn't it? or is it just going to show you random websites? Here's a new website I found.

Um, you know, I think it's a little strange, isn't it? Like I' I'd be interested to see what it does and how it behaves because having had some experience of using some of the built-in tools like Gemini for example into all of the Google stuff I don't make any use of it because this the output that you get from uh you know trying to get it to support you in writing a document I don't find it useful at all. I'd actually rather write it myself. um you know but I do use AI on a daily basis within cursor and I you know the other advantage of it being within cursor is that because I pay for the pro my code is supposedly not shared anywhere and I would worry in the same way that you know stuff that I was doing on the internet was shared elsewhere and I don't mean that in necessarily nefarious way I mean like if you're doing something that's creative or you know whether that's writing a story if you're a writer or writing a document that if you're a business person, I don't want that shared anywhere else.

I don't want that to be fed into training some somebody else's ideas because the, you know, I think we're still at a stage and I think we will be for a long time where AI is still regurgitating stuff. It's not inherently creative. Um, so I would be very interested to see how it works and if it changes how you browse, but I don't think I would necessarily see any immediate benefit that I would gain. Yeah, I'm struggling with it. But in the same vein, Opera also announced their AI browser that they're going with right now, Opera Neon. And again, this is supposed to understand everything and help you with tasks. This has got like a it's like a three-step program, right?

You've got chat, which is your classic chat if you want it to help you out with certain things, answers, and stuff like that. This is really weird, actually. This is kind of strange. Um, which, you know, I think we're all very familiar with chat bots. You've got do, which is actually um like building things like and No, no, so sorry I'm getting confused. This one's an agent. And are you familiar with agents? Have you been digging into the world of AI agents, Chris? Yeah, I've built a few. Built a few AI agents. Jesus Christ. Oh, yeah. Um, as in you've used platforms to build them, right? Um well if you if the platform you're talking to build them is like digital ocean where we're actually doing rag and augmentation and things like that you know that's actually what we've been doing of um we did a project relatively recently which was supported by innovate UK so it was my company anything and in good company we did a partner a partnered um piece of work on there which actually was about training agents on company data uh to give sort of more contextaware stuff.

So working through different workflows. So we have done quite a few agents. There's definitely going to be a um there's definitely use in agents for sure. Absolutely. I mean like Yeah. No, you're you're like literally as close to the as close to the the system as as you can if you're deploying your own own agents. Um I thought you might have been using like N I think it's called or something like that or Well, we have used NA. We have we have used NA10 for creating workflows but not actually to um to to actually not actually integrating with the agents at that stage. Literally just using it to go and gather data from various sources and drop it into um storage that basically forms a knowledge base for for an agent which is then trained elsewhere trained outside of um NA10.

So we haven't used nen in that that scenar in that way. Uh but we have used it as part of certain flows just to have a workflow engine that can run repeatedly on a certain schedule. And what models are you using for your for your agent? Well, we've tried a few um and I think it's interesting that you know the ones that are more sort of readily available are the likes of Llama and um it's generally not great. It's all right. Um the quick things I find well actually the the smaller models are but yeah. Yeah. Yeah. But now that we're getting to the point of um putting some of that stuff into production um we are experimenting with slightly more advanced stuff that feels a little bit more natural.

Um but what's quite nice is because we've got the workflows that generate the uh generate the knowledge bases actually swapping out the agent still providing it the same context. Um, but swapping out the agent for a different model um, is relatively simple and it's just really trying to figure out the costs for that. So, you know, I think we're finding that certain models are better at doing certain things. I mean, I quite like Sonnet um, for coding. Um, I think some of the chat GPT ones are are better for for language. Um, so yeah, I think really it's it's just about finding which is the right model for the right scenario. But Llama's definitely good to sort of test things out, but I don't really feel like it's the sort of thing you'd want in production cuz it tends to hallucinate quite a lot.

It's still got that, you know, that old school hallucination where you ask it to retrieve something from its data source and it goes, here's something I've just made up. I asked you not to do that. Explicitly ask you not to do that. Yeah. Exactly. Yeah. Yeah. Are you using open router for that or are you just sticking to models that uh adopt the open AI uh standard? Um mostly I think they're just following the open AI standard. Yeah. Yeah. I think uh it's only really Google Wow. Yeah. Only really Google doesn't. Um, I'm building an AI app and I'm thinking of just moving it over to open router just because it can just swap out models using the same sort of, you know, standard and and whatever.

But um, anyway, we're getting sidetracked. Um, so yeah, this is more of an agent thing and this is so I'm and we'll get into Perpexity Labs which has been launched last week. This is more of a um an agent that will go off and do several things at once. And I had a video uh I think a video's just gone out today actually. So uh yeah um or no yesterday whatever. And then make this is a really weird addition to a browser. This is a vibe coding app where you just tell it to go build you a website or a game or something like this. And remember this is all inside of a browser. So I don't really it this feels like a bit of a personality.

What's the word I'm looking for? Just um just doesn't know where it fits in the world. It's kind of like throwing everything at the wall and seeing what sticks. Um, but it is interesting. Well, I mean, are people still using Opera? I mean, that's a I they are trying. They're damned us, but I don't know. Well, I don't even know what um are they what are they Chromium based? Are they like I don't even know what they use. Um I don't know. I think I think a lot of browsers moved towards Chromium a few years ago, didn't they? Yeah. Wow. There we go. There's Wow. I don't know. So, look, I've got to log in.

And I was going to use my little chat app to uh find out if Opera is Chromium based. So anyway, this is the this is the kind of browser wars that are kind of going on. They're doing interesting things and I'm I want one of them to release something like because I'm struggling to put, you know, put my head get my head around the concept of having a a AI wrapping my wrapping my browser world. I think it's still a little bit like when all of the IoT stuff kicked off, which really hasn't still hasn't settled. You know, again, I spent some IoT is something I spend quite a bit of time doing, as you know, um from the ridiculous automation in my house.

But, um it's it's kind of it's a solution looking for a problem, if you know what I mean. And it still feels a little bit like that with AI. you know, everything just gets an AI sticker slapped on it in the same way as, you know, 10 years ago everything had a big data sticker slapped on it. Um, which obviously is actually the precursor to AI if you think about it, I suppose, or if you follow that thread. Um, so yeah, I think it's just one of those things at the moment it's got to have it's got to have AI in it. You know, in fact, actually a great way of looking at this from an indication perspective is in terms of the funding that is available at the moment to startups um and trying to get funding from Innovate UK for example or UKRI, you have to have it.

It has to have AI in it otherwise you're not going to get your your project is not going to get funded. And the same goes for um for investors for for venture capital uh investment. If you do not have AI as part of your project, you're not getting funded. So, it's one of those things at the moment. It is the buzzword and if you're going without that, then you're going to struggle, I think. Yeah. Um, it just go back to what we were saying in the preamble before the show. It's like, you know, Apple always hid the fact that they're using AI and then they started losing investor and and shareholder um trust. So, they have to start revealing, oh, we're actually using AI for this and we're actually using AI for that.

And it's so strange to people for people to really care about the underlying technology. Do you know what I mean? So I think there's always been a bit of that. You know, if you go back to some, you know, even if you go back as far as the 90s when things were starting to go online, you had E stuff or I stuff, you know, there's always been a little bit of that. Yeah. Yeah. Yeah. I mean, it's technology is so pervasive of it's just it's all around us. So I think whether people are technophobes or not, it's just they're being bombarded with technology just because it's so important. So yeah, but um anyway to the you've just updated cursor finally become a big boy.

You've become a big boy now. One up to uh up to version one. Absolutely. Yeah. So uh there was a few things I made a little Tik Tok on it yesterday. It was quite fun. There's um bugbot which basically as you can see it just here um I'm not quite sure how it works exactly but you can tag cursor inside of GitHub as it's doing you can see right now on the old screen share you can tag and and click okay so you click on fixing cursor and then it just opens up and whatever because I guess VS Code has a GitHub integration right uh you can do pull requests and things like that so I'm assuming that's the sort of technology it's utilizing but I guess it can just pick up the the bug that's in a in a pull request and then you can fix it.

Um well that's interesting because there was actually um what was the tool I'm just having a look through my git to see if I can find it but there was a tool um an AI pull request tool that's been that I saw at London Tech Week uh last year actually uh what's it called? Sorcery AI. There you go. Um so that does your pull request reviews for you. Um, and it's relatively good actually, but I guess it's uh curs's move to try and uh copy that I guess to a certain degree. There's a few I mean like obviously you can tag Claude coat. Is this Does this look familiar? Um I think probably Yes, it is.

No, the box is that little box is the same that little logo. Yeah, that classic box. Okay. Yeah, cuz like um cursor code can do this. You can tag cursor in your pull request and it will fix it. Obviously, jewels as well is kind of like a thing, but there's a few kind of doing this where you just tag them in a pull request or you tag them in an issue. They'll go off and fix it and create a pull. I guess you tag them in an issue and then you they go off and fix it in the pull request. So, um, yeah, there's a few of these and I I love that idea of just being, you know, on on the on the bus and being like, "Oh crap, there's there's something.

I noticed an issue, report the issue, get the AI to start with it, and then you pull it down in cursor later on and fix the damn thing." So, I quite I quite like that this this idea. I've not had much experience using them yet, but it's I I tell you when it's I tell you when it's useful to use something like sorcery. So if you've got a relatively large team and you're doing the pull request um it you know pull requests are great if you are um tool well tools like uh sonar I've always used tools like sonar um because at least then if you've got junior members of your team then they can sort of fight with sonar or fight with the static analysis or in this case fight with sorcery until you've concocted a PR that is actually ready for view and has, you know, worked out all of the security bugs or worked out any of the other bugs that it's able to identify um and gets to a certain bar before you then have a human or a, you know, senior code reviewer actually reviewing something and saying, "Yeah, that's ready to go."

So, I think it's great in that respect. And this is actually, I think, what AI is going to be really good at. It's going to be good at going through its its knowledge base of what good code should look like and determining whether there is um uh any issues. More like more of an advanced static analysis tool. And I must admit I use cursor regularly to review my PRs and say right is this ready to is this ready to go? Do you notice anything in this? Have you spotted any duplication that I might have left behind? you know, if especially if you're doing quite a large uh release, it's it's particularly useful for that. Also very unnerving because when you're trusted, you've got to really trust in an AI when you're close to a release and you're saying, "Is there any issues with it?"

But um well, if you're doing a small increment, a small feature, you know, you're just checking with small features. You're not checking for I'm not suggesting that you you code for six months and then ask the AI if that looks good. Yeah. Yeah. Yeah. Yeah. Strong. Um, are you uh because anything like how are you using AI to to build that? Presumably it was built without AI. Um, but are you kind of encouraging the guys to now use AI to add things and change things? Uh, absolutely. Yeah. But I think, you know, we're a small team now. Um, and we don't do a huge amount in AI actually. Sometimes we will prototype in it. Um, but because of the level of what we're trying to achieve, um, we often don't use that much AI because we're trying to make sure that stuff is, um, particularly robust.

So, you know, we're dealing with other people's data, I suppose, because in one respect, you could use us over Superbase. So you want to make sure that you've got some really strong um uh security con constraints in there I suppose to make sure that everything is going to work exactly as you expect. But it's not to say we don't use it. Um because we do use it as we go and like I said largely for prototyping and also you know there will be an actual AI tool in there further down the line um because we've been experimenting with it already that where you will actually be able to build and design your data models um by asking it to do that for you and it will do it for you.

Um it's just that will be in a future version. I was going to say is it is it live yet? So, we're using it well, we're using it with a number of customers, but we're about ready to we're getting to the point next sort of month or so where we're going to be able to actually release that as a public um platform for other people to just start their own instances. I have a banner ready to go that says anything.com. Should I put the banner up or not? It's anything.uk. So, I would change where is it? Go on. There we go. Add show. There it is. Anything.uk. Is that it? That is right. Yeah. Yeah.

So, you can go and have a look at anything.uk. Um, we've got some cool stuff in there and there's going to be some a lot more to come. So, um yeah, next couple of weeks I I we can I can come back if if you'll let me and I can tell you about it then when it's ready for people to just go and start their instances. Yeah, let's do it. Right. Anyway, so you mentioned something when it comes to cursor that I think you might find really helpful. I mean, claw code introduced me to this idea of memories. We'll go back up. Don't worry, folks. We'll go back up. It's just because we mentioned this.

So are you are you aware of memory? Have you are you aware of claw code claw.md files or anything like this? So uh I have used MD files and I have used um the cursor rules as well which is quite useful. Um it does tend to ignore it on occasion which is a pain. Um but I think that if it can start to do memories that's going to be quite useful. Um, I think you know the the um I've not had chance to try it in cursor as yet. I think the Well, you just me just before this call, so I don't doubt you have. Yeah, but we'll have a look this afternoon for sure.

Yeah, unfortunately after this call, but I think if it can start to remember stuff, it's going to be helpful. So, so what I like, there's a few reasons why I like this. I mean, you could you can go as broad or as narrow as you want. You can you can set up a memor I'm speaking from claude code perspective here but this is basically the same thing you can do it per folder so you can say this is the roots folder and this is where all the pages live and you know all that kind of stuff and you can slowly work out to a project you can do you can you can put your um let's say claude code uh chord file in your your machine's kind of route so it's how you like to write code or you can put it in your organiza you know it sort of broadens out and what I've noticed cuz I'm using cord code at the moment is that it tends to try and update that memory file or that cord file if it's in a folder it starts to update it and and modify it so it m it maintains itself.

So I'm hoping that this is a similar sort of thing and that if you have got it in a folder or you have got it somewhere that it's constantly updating this memory uh to changes that you're making. Maybe maybe a file doesn't do the same thing as what it once did or a folder didn't serve the same per or whatever it is. So, I'm really excited about memories. I've alluded to it in a video that I released last week of I think this is one of the big this is before cursor released it to be honest, but uh memories and task management are two big things I think that are going to help us start to step back from um you know hand coding things and have more trust in AI coding things.

I think those two things are going to be huge. We're going to be writing um what do they call it? Product requirement documents as opposed to, you know, make this function do that. But that's just my see I wouldn't necessarily suggest that it's going to allow you to be more handsoff but perhaps it's going to allow you to be more consistent because I still I still think that you can't let it just get away with everything because quite frequently it will change the architecture you know especially if you are if you are doing something other than say um you know more standard things like I'm I'm coding in next and tailwind like if you start going, well, actually, I'm going to use um radics and I'm going to use uh uh even just Vite, you know, I'm going to do something slightly different.

Then occasionally it's going to try and sneak in a bit of tailwind there because actually there's more tailwind on the internet than there is on there is radics. So, you know, you're always going to have that sort of slight battle with it, I think. So, if that starts to solve that problem, then it I don't think it's going to let you be hands off, but it's going to um at least allow it to be more consistent. And I think that that's going to be useful. Yeah. I hear a bit of doubt in your voice, Chris. I don't think you're fully AI pill just yet. You want to be hands on? No, I'm I'm not. I want to be totally hands on No.

Well, do you know what? I I I was listening to Radio 4 the other day and there was a person on it and she said I I do every day and there was a person on it and she said, "Look, I don't want the AI uh to be um to be doing the painting and doing the writing so that I can do my dishes and my laundry. I want the AI to be doing my dishes and my laundry so I can do painting and writing." And I think it's kind of the same in this sort of thing. I don't want it to do the creative stuff and it's actually not very good at that in my opinion.

I want it to do the boilerplate and make things really simple and easy for me. Um, but I also and I like I said I use it every day but you can't I just don't think you can let it just go on its own and I think as as especially as code bases get larger and more complex uh bugs start to creep in. And the real concern as well, which is why I use it to sort of check a pull request is that it doesn't seem to be very good at self-reing and you know, so it can't code and then review itself as well as it really should do. And that's where security feature security loopholes can creep in.

And I think we're we're potentially on the verge of if everybody carries on with Vibe coding, people are going to start getting sued because bank details have been leaked. you know, it feels like it's that sort of direction. Yeah, I think my my handsoff approach is very much like look, you have a good first start, but there's nothing I vibe code or you know, you know, when I am hands off, my my definition of hands off, I still need to look at that code. I still need to know and and understand it and make sure it's doing the thing that I half expected it to do. Um, I also want to learn. It's a great way to learn, but um yeah, I don't know.

Well, I think I think we it's really interesting where we're going with software development um with with AI. I I'm I'm of belief that uh seniors and and mid-levels are going to be more like more reviewing and orchestrating co code rather than actually writing it. They're going to be the ones checking that everything's good and and again just being more pedantic with what what goes in and just not not blindly throwing code in. That's that's, you know, junior level development. But anyway, um we've also got background agents, which is pretty cool. I love this idea of like, you know, you write something. Well, actually, it's something I've um not had a tremendous amount of experience in, but you can uh remote coding agent, so you can set off agents to to run and do things on when as you write code, which is which is pretty handy.

Um, have you had much experience with background agents at all? Because I think that No, I haven't. I was trying to work out exactly what it what it did. Um, well, I'm guessing it's kind of like a tool call that runs um, you know, with with a because you can prompt it then you can actually suggest that it does something uh to run in the background. So, it could be just, you know, go and do this thing whilst I code this other thing, you know. Well, I did see there was some stuff in there about parallelization, and I do think that's, you know, going to be useful because that's one of the things that's probably most annoying about doing coding with AI at the moment is you spend you you waste time waiting for it to finish something.

Yeah, exactly. I think that this is um well, I think this is more so this could be more so like they showed a demonstration or an integration with Slack, you know. Um so, I don't know. It's um it's an improvement. We'll just trust them with their word. Jupyter notebooks. I mean, I don't write Python, but I think the thing is um I saw that VS Code released a released like a tweet or something like that and like talking about their Jupiter notebook integration. So, I think they're like, "Don't forget us. Don't forget us. We we we've always done Jupyter notebook." And it was like a day after the um cursor announced their Jupyter Notebook integration.

They see that's always that's going to be an interesting thing. that isn't it because cursor is VS Code so you know everybody's jump shipped to to cursor where does that leave VS Code well they I mean did you see they open source their their copilot agent um actually people getting confused they think that that that they've just given away C-pilot for free but it's not it's not copilot they're giving away for free it's the it's the architecture it's the it's the mechanism it's the it's the chatbot on the side to go off send messages to a thing, but you still need you still need to integrate. It's you you still need to pay for o like um co-pilot.

Microsoft's you overuse of co-pilot is just too confusing. Um but copilot the actual coding assistant that reads your code and then try you know interprets it and all the rest of it. You still need to pay for that but the the the framework that runs all that that's been open sourced. So you know one of the biggest aspects that um of the reason why these coding these cursor the wind surfs um and you know plugins like clin and ru code that's the reason why they got created because they had you had to build upon um or you had to build an extension from VS code but now the fact that they've open source that leaves them in a in an interesting spot.

Um, but yeah, this is what I think. How I mean, I'm going to ask you all the time, aren't I, Chris? Are you using this? Are you using that oneclick installation of MCP servers? I think this is going to be incred. This is like the app store of MCP servers. I don't think people are quite quite understand how great MCP servers can be. I watched a video last week of a guy who basically created a YouTube video through MCP servers integrated with his I mean Claude the app Claude had it all along. I think um uh cursor has had it for a while but this oneclick access just makes it more accessible. He just created a YouTube video just from a prompt.

It kind of generated a script. It generated storyboards. It created images. it it blended between those images using foul generated a voice over all through like natural language and describing it in a thing. So yeah, have you had much experience with MCP servers? Um the closest I suppose is obviously the experimentation we're doing ourselves with with with anything but I think the first time I was really impressed with it was the was the integration with GitHub um in cursor where you know I had a pack I had a I had two apps one that um consumed a package from another and I needed to upgrade one of the packages and get the app to consume the new package and I I didn't know how successful it was going to be but I said look, I've got I've got this task to do.

And he went, "Right, okay. I'll go and do that." Um, submitted a PR, merged it in for me for the new app, published the package, and then updated the the new app to consume the old package to to consume the new package. And it it just worked. I was like, that's actually, you know, pretty damn impressive that it was even able to raise the PR and um and and merge it in, which was great. So, I was quite impressed. Um, so it's nice when things like that do just work, but um, yeah, I haven't experimented with some of the more wild um, MCP servers, but uh, but yeah, when it when it just works, it's nice.

Yeah, I for the life of me can't get I haven't done it in cursor. I haven't tried it in cursor yet. And obviously this one click is going to make sure it bloody works. But I've tried setting up one in in in Claude. I just can't get it to work. I don't know. It might be a Node.js JS issue, but um slightly annoying. But yeah, I think the again the one-click access again get people like me who don't know what well yeah don't know what they're doing think it's going to be really really cool and then I think they're just there's just going to be explosion of loads of MCP servers that I mean there's the app store there notion um Figma and all the likes of it.

So I think this is going to be really good. Richard chat response which is just a just a more interesting I think I show graphs and tables and things like that. Um, always welcome. Nice little update to the settings dashboard. Cursor one, it's become a it's it's grown up. It's become um a teenager now, which is pretty cool. Um, uh, have you Gemini uh, goes global. So, Gemini Ultra, which gave access to Have you seen like the V3 stuff that was been released? It's crazy. I haven't. I mean, I've been using a little bit of Gemini here and there, and I have found as well that um using Gemini for certain languages actually like Flutter because it's Google based does seem to actually know its stuff a little better.

Um, which I thought was an interesting thing, but uh but no, not not not particularly familiar with the VO stuff. So, I mean, Google, it was it's a $250 plan basically that included obviously Gemini, the chat apps with uh the advanced um uh models. Uh but the V3 stuff, which is just some crazy video generation video, I've just been I'm just trying to work out whether this I I was just sending my friend some really I'll link it somewhere of these. Um, my phone's probably going to blast out a bunch of audio now. Um, these uh what do they call it? The Sasquatch. And uh I've got I've got a feeling it's it's I'm going to I don't know whether you can see that, but it's this Australian It's not going to focus, is it?

And it's black and white, so it's not But this uh I'm not going to make you watch that, but it's it's these funny videos with these Sasquatches generated. And I'm wondering if it's V3 because it's so convincing and it looks so good. Um, but the point is this all this was like in America only. So they've rolled it out globally and it's 234 I think. Yeah, 234 um dollars a month which is really cool. The V3 stuff is available on the pro plan, but which is only 20 bucks or something like that, but you you get very limited usage. Like even under on the $250 plan, you got limited usage. But V I think it's 10 10 uses a month was it or something?

10 videos a month on the Ultra or the Pro. Uh I don't know. I saw it in this article. I think it was 10 uses a month. You get it sounds like that's even the Ultra, you know, it's so it's that intensive. Um something really interesting actually that I found out about reading about this is this synth ID watermark technology. So what this is doing it's Google deep minds um created this identification system to obviously let obviously Google have created this ecosystem but they can tell whether something's been generated with AI and they mentioned something about cropping and filtering and I'm guessing it's just these little things here or there that happen to let you know it's like a key I guess that lets you know that it's been generated.

So, this is big business and I think this will be the one of the most important things going forward to know if like you're actually watching a politician do something real or whether it's a net, you know, deep fake or whatever. Well, I think that's interesting because I suppose it's a little bit like DRM technology. Um, the digital right management stuff. Um the the thing that struck me with that is if you were to say do a pass over a video, you know, even if it's just like um what providing a lot I suppose over a whole video, is that then going to watermark it and say that this is AI? I don't think it's I think it's invisible watermarking.

So I think this is the whole point is that it's unre like you can't tell. It doesn't destroy the experience. Oh no, I know. Yeah. Yeah, but what I mean is if you say something's been, you know, filmed with real actors, it's been recorded, it's been edited, it's had its audio done, and then as a last thing you pass it through some sort of filter. Is that does at that point the watermark get embedded and then you're now it's now tagged forever as AI? Good point. I mean like because obviously again just before the call I mentioned about the AI exists on many different layers of what we do right so you're set you're suggesting that there's an AI L however the editing the audio the the acting in of itself was not AI generated but now it's just like oh okay AI generated because you use an AI L is that what you're suggesting yeah it's been tagged you know and watermarks as AI or having AI involvement when actually it's only that AI involvement was very minimal potentially.

Yeah. Yeah. And then I guess there's a whole there's a whole ecosystem around like well maybe that's just a prompt to the user to suggest or a prompt to the system to say it's been AI generated and it's up to the the artists or the creators to actually be transparent about where it's used because I think Mr. Beast went through a whole ordeal where to make their videos more dramatic they add you know CGI and explosions and do all that cuz you know it's a kid show and that's what happens but people got very upset. I mean, it was, to be fair, I think it was on one of his kind of um where he helps out people and stuff like that.

He was demonstrating what a school looked like in Africa before they helped rebuild it or provide water to it or something. Anyway, point is is uh yeah, I think people are like they feel cheated if if an ounce of it is is AI generated. Um but yeah, yet to be seen. I think it's uh I think I think Synth ID is very much apparently they've um watermarked over 10 billion pieces of content already with it. So I mean I know that's small in the grand scheme of the internet but it's a lot. It's a fair amount. It's a lot. And then on the subject of um of Gemini Ultra and stuff like that is uh Pro Deep Think as well you get with Ultra which is a hybrid model that's supposed to be just amazing at everything.

It's super. It takes the um deep thinking and speed and rolls them into one as well, which uh yeah, we've spoken about on the show before, but like that's kind of what you get. Apparently, it's good for coding as well. So, that's pretty cool. Um and also then back to back to web um cord pro again. You you you use cord to code? I do. Yeah. Um but within cursor so you know predominantly um Claude Sonet 4. Yeah. At the moment. Yeah. Yeah. So I bought the max plan last week because I wanted to use uh Claude code. I wanted to give Claw code a go. Uh have you used core code? I haven't actually.

Um, I did read a review um comparing cursor and wind surf and cord code and I made the decision to just stick with um with cursor to be honest. Well, I can tell you it's not the video that's just it wasn't my review that I just launched because you've been on the show. It's uh that just went out today because that's exactly what I did. I just compared the two. Um my my opinion my my opinion is although it's great unless you're like fully immersed in the ecosystem of Claude you know um and arguably and again this we'll get into this now that um Gemini 2.5 Pro has been updated again and Gemini 2.5 Pro is like you know a a fraction of the price of of chord and arguably I mean it's top of the uh sweet benchmarks I think it is as well so it's doing really really Well, so unless you're fully immersed in the in the chord um ecosystem, then you might as well just stick to just experimenting with it, using the Max plan every so often.

But the news here is is that they've now released it on the pro plan, which uh it wasn't available. You had to get this $100 plan to get access to Claude Code. So now for 17 bucks, you can get probably a reasonable reasonable amount of access. So again, what's your bill like? Are you use actually are you using max mode at all? No, I'm not actually. Okay. Um and I I think I'm worried that it will just spiral out of control very quickly because especially that's exactly Yeah, that's exactly why these plans exist. Yeah, because you can get to the um you know where it'll tool chain like 25 commands. Yeah. And you know if you accidentally turn your back on it, you know, you think you think it's doing one small thing, you turn your back on it and it's, you know, done everything and decided to do much much more.

Um, you know, that's the concern really, I suppose, where it gets stuck in a loop and it's racking up a bill without you really being aware that it's doing that. Yeah. Yeah. No, exactly. And that's why these this plan got released because it's pretty much, you know, inf it's not infinite, but it's it's good enough. It resets every 5 hours and it's got a very very generous limit. But the max mode, yeah, as you say, that's using the API. You're paying for your cursor and then you're adding the API usage of Claude on top of that. Whereas this again, you wouldn't you would you you will watch my video later on, won't you? But you clo you close your cursor, you close your close your cursor chat, and then you you sit within uh claw code then and then you've pretty much got an unlimited usage of like uh Sonnet and Opus.

But um yeah, it's just nice to know they've released on the 17. Are you finding Claude Code as good, better, you know, how's it still still making my mind up? Um because you leave it in Opus because Opus is a is a uh what do you call it? A hybrid model. And it's supposed to work out whether it can drop down to Sonnet level or whether it needs to up its own game and go to deep thinking mode in Opus because Opus sort of overtook Sonnet. It was Opus was the one kind of Yeah. So it is good. I've I've I built a WordPress plugin today with it. Um I launched I'm build I'm working on a free version of my chat app with it.

It seems good. It feels I don't know why it might must be a shallow ego thing, but it feels nice knowing that I've paid a hundred bucks for this one tool and I'm just coding with it in terminal. There's something I get I feel like a Vim user, you know, I've got the I've got the Vim snobbery. Yeah. And the Vim snobbery is real. Yeah. Yeah. What? You use your mouse? What? Yeah. A gooey. Especially when you get people use T-M. Yeah. Jesus. So, um, anyway, from a code perspective though, it's it is I don't know how many tokens are wasted by being like, you know what, you're right. That wasn't supposed to go there and things like this, but it's it's fine.

Um, I'm I'm going to go straight into I've got a month of this now. I might as well use it. I'm going to go straight into Gemini uh Pro 2.5 Pro once this does because for the fraction of the price, you know, even in I I don't know whether um I haven't got cursor up. Um I don't know whether max mode is available in in Gemini Pro, but the point is like I think arguably Gemini Pro is just so much more value for money, you know. Um, but all that being said, you can try it out for I was because if you went on to max mode, I've heard of people in max mode in cursor, not to be confused with max plan here, but max mode in cursor running up 200 300 buck bills on top of their cursor plan, you know, just using the max mode of GE.

they're happy to do it because effectively they've they've they're employing um two or three junior devs to you know slog away at the code or whatever it is. So I do think it's worth it but um yeah so that was all introduced because of that. The the other thing is at the moment if you're using uh claude for set within uh cursor the requests are at 0.5 times. So there is actually a discount at the moment. Okay. But is that so the conf the pricing is so confusing. Does is that in not in max mode though? Is that not max? That's not in max mode. So you can up it to max mode. Um so do you get a bill then on top of your c because I've never got a bill on top of my cursor bill.

Uh I can have a little look or find out now. you get you're going to get live shock now as I go how much cuz or is it a theoretical price cap to know whether you've capped out your 20 bucks because presumably you pay for cursor right? Yes I do. So I'm on the I'm on the pro plan and then there is a a sort of premium. It used to be premium or pro um level that you could add for like premium requests. I would say premium requests. Um, so let's see where we're at here. H, there we go. So you can share your screen if you're confident enough, but um, probably not. So there's a $60 max spending limit that I've set.

Um, so it re it's halfway through a month now and we're $3 in. Um, and that's with So you are paying on top. Yeah, we are paying on top, but that's with a team of three at the moment, and it's only one of us that spent that $316, which is interesting. It's not me. There you go. Oh, okay. I thought it was going to be you. I was like, "Oh, who's not picking up the slack?" It's not me. It's not me. Um, but yeah, I think that's literally on which model you're choosing. And so I think much like we were having the conversation before um I would rather um just have a model that's not going to make mistakes um and not invent things and not hallucinate and then I would rather sort of code with it if that makes sense.

I like to treat it a little bit like a junior developer to be honest of getting it to do the hard work the hard sort of leg work for me. Um, but I will give the architectural design because any, you know, haven't really used the max modes to to full extent, but I just don't necessarily think they've got the architecture or the creativity required yet. Yeah. Yeah. So, I'm going to be producing a video which um, you know, the long and short of it is I think that the max mode, Opus, the these expensive, these deep thinking modes, I think this is where you start your big feature. This is where you plan out your big feature and say, "I want to do this."

And you you give it a product requirements document to plan that feature out and you might start to shape that a little bit with that max mode and then you drop down to something like Sonic or something lighter to start then chipping away at that feature. So you just sort you just start shrinking down your scope starting out with the with with the big version sort of thing. But um yeah, I think that's an ideal situation. And I think you've it's you've got to build trust. It's got to build trust, right, to actually get the get the get the job done and and whatever. But maybe not on a on a live app, maybe on like a hobby project or something.

Um how long have we got you for today, Chris? Are you shooting off in five? Uh yeah, I can I think we can wrap it up if in a bit if you want. Um, we got what? We got Plexity, we've got Sakana, and we've got 11 Labs three to go. So, um, yeah, three more to to waffle on. Well, I'm kind of I'm kind of interested in uh in in the 11 Labs one for sure. Cool. All right, let's do 11 Labs now. And I can keep going with um if you have to shoot off. There you go. Compromise. You got to have good compromise in a relationship. This is the thing. So, do have you seen this?

I have. I have. Do you want you want to chat about it? Yeah, I think it's kind of interesting, right? So, it's it's definitely much more expressive. Um, it it's it's kind of cool that it can do more accents and and uh and and and voices and expressions and all of that sort of stuff. And um actually there's one project that I work with that uses uh a lot of audio content which largely uses a lot of 11 Lab stuff and I must I didn't actually notice they were doing that. So because I know I looked Oh, what you thought it was an actual person speaking? I hadn't paid a huge amount of attention to what was actually being said, but it hadn't twigged with me that it was AI until I was told that it was AI.

Wow. Um which I think is quite good. But um yeah, I don't know the other thing from listening to the you know, I've gone through the their little video clip as well of of what's in V3. Yeah, there is a certain element of it. It's very impressive how expressive it is, but it is almost a little soulless. Um, so yeah, you know, you can kind of tell to a to a degree because I'm not saying you can't because clearly I've been fooled by it before, right? So fool me once before we tried to. Yeah, exactly. You can go on forever, can't you? Fool me three times. Um, yeah. No, the the um Yeah, there is something about it where it's very convincing.

Mhm. But it does seem to lack a little bit of soul. And actually, I was talking to someone about this sort of thing earlier. It feels almost a little bit like, you know, when when drum loops started to exist and now, you know, you do that by sampling each of the drums and then you stitch them together. And now there's pretty much every um you know, audio tool that you go into, you can write, you can do your own beats and you can create something that sounds or approximates a drum kit. Um, and obviously there's certain things where it can pre-program that for you and it will sound like a drum kit, but it's not going to sound like a drummer, you know, and I think that's the difference.

Like, it's very, very convincing and it will satisfy an awful lot of people and convince a hell of a lot of people that it's a real thing and it's, you know, real music to go back to the the other thing. Um, but there's something different in comparison to a live recording. Um, it's always going to be an approximation to a human, I think. Um, so very, very good. Sounds great. Um, great for some brilliant purposes, but thumbs up to the 11 Labs crew, but you lack soul. It just feels it feels like there's something missing, right? You know, to me, but it's very, very good. It's very impressive. I'm guessing this is like the prompt, the AI thing.

I mean like I I was actually in shock in awe like that some of these like so which one might it be? I don't know. It just sort of takes into account the whole entire context. And so when laughing let's I'm just making this up right now. Okay, here's one here in the context of this one. Like let's say laughing is followed by hysterical laughing. Laughing in this context might mean that it's more aggressive than laughing without hysterical laughing. Again, I don't know. I'm just guessing. So, it just felt like there was I you could put laughing in different areas and it would respond differently based on its context. So, I I was pleasantly surprised.

saying all that, you know, you were probably hyperfocused on the fact that you were just reading text on a screen and and watching the words and and listening to the audio. Whereas realistically, and again going back to this idea, this guy who used tool calls or or MCP servers to create a um a YouTube video. He used 11 Labs and it was in the context of a video like things happening on the screen, you know, a conversation or something like that. So maybe you're extra sensitive because they were so brazen enough to just present you with the audio and a simple bit of text. But it's definitely uh this definitely got my cogs turning on what I can use it for.

It's possible. I mean the other thing that's interesting about it is you know I listen to a lot of audio books. I listen to a lot of podcasts and you know there's no there's no substitute for a really great um narrator. Um, you know, it's always going to be compelling if there's a great narrator. However, recently the current book I'm reading listening to, you never know what's the right way to say that. The current book that I'm listening to, I'm reading a book. Oh, yeah. Well, well, it's an audio book. Well, it's not I'm not read. But anyway, the current one that I'm listening to um is a great book and I'm I'm actually really enjoying the book.

Um, it's a it's a Winston Churchill book. However, the um the narrator is awful. And I didn't realize because as I was it's important to know that it's a Winston Churchill book because as you read the a you know you go through the audible preview and you go yeah this is fine because the guy's reading the uh reading the the preface and the the introduction to the book and then you get into the actual book once you've purchased it and find out that the guy actually does most of the book with a very bad Winston Churchill impression which is very disappointing to now if I could have re rerun that and gone actually take out the Winston Churchill impression.

That would be fantastic. Yeah. You know, in that instance, I wouldn't mind that being soulless. It's not going to be, you know, uh it's not going to be a substitute to, I don't know, listening to something that's read by Stephven Fry or Tim Curry or someone like someone, you know, a really expressive actor or actress that's going to do a great job. Um, but it would be nice if you are uh subjected to someone's poor quality impression to be able to go, do you know what? Stop that, silly. Yeah, please just stop. Yeah. And we had a brief conversation about this a few weeks ago. Um, it I think creativity in this respect, I think you've got to have you've got to be a creative like AI is there.

You can't just eliminate AI just because you're doing something creative or whatever. I think you need to understand where its strengths and weaknesses lie and you you just have a have to have a bit of a moral compass when it comes to how you use the AI sort of thing. Like you know if you're if you're a massive like movie like like Fox or some massive like uh company and you're you're choosing to use AI just because it's quicker and cheaper. I I think that that is a poor use of creativity and a poor use of AI. whereas someone like you or I, you know, just um trying to trying to make things work, trying to grow their company.

Um I think these platforms just inspire me so much. And but you need to know where that line is should be drawn when it comes to okay, am I just am I taking the piss now? I should I actually just go out and hire Steven Fry or you know rather than using his AI generated voice or things like that, you know? So, it's a tricky one and everyone's going to have a different answer to it because everyone's moral compasses lie in different areas. Every everyone's understanding and appreciation for creativity lies in different areas. But it's an interesting dilemma that you've had because it's like you it's actually nothing to do with you're not taking away anyone's uh like creative output.

Maybe they wanted that to sound like a crap version of Winston Churchill, but uh I'm sure they wanted it to sound like a very good version of Winston Churchill. unfortunately didn't have the budget. Yeah. But yeah, 11 Labs, it looks cool. I'm I'm I'm buzzing with ideas to think about it because again, solo entrepreneur, whatever you call it, solopreneur, someone I want I would love to do some little adverts and things like that, which can allow me to add value in in other areas of my life and things like that. So, I'm I'm defin I just get I've described it before, I'm like a kid in a candy shop when it comes to all these uh these AI tools.

But I think that's what you should use it for. It's it's it's a creative thing and you know, well, it's a creative shortcut, I suppose. And if you if you can if you can use it as a tool for your own creativity, then I think that's great. But I don't think it is a substitute for creativity, you know, and I think that's the, you know, it is that same concept. I I do want AI to do the manual labor for me so that I can be creative. I don't want it to do the creative stuff for me. Yeah. Yeah, for sure. Um, and you know, Yeah, I totally agree. And the 11 Labs is an alpha, so I don't even think you can access it even through the API stuff yet.

But I was pleasantly surprised at the pricing. I was pleasantly surprised they are blatantly trying to get devs on board now because they got you. I in fact, this seems to have changed, but when I checked it before the call, this had a lot of emphasis. Oh, I guess it's the API access. M um I think they really want devs to use it given that this is you know free API. I mean I guess it's it scales upwards from this point onwards but like yeah um when once once V3 is out I'd like to play around the API for sure. Yeah absolutely. I think that's the next step for me is that you know I've seen little clips of people using um you know chat with voice.

Um, I've not had chance to do that yet, but I think that does feel like it'll be the next step of actually having something read back to you. Um, be able to have a conversation with it. You know, I've been doing a lot of driving recently, and if I could have had a tool in my car connected to my Apple CarPlay that would allow me to have a conversation with it and get it to write a document or take down ideas or take down notes, that would be such a great tool. I'd be able to make so much more use of the time that I've got driving um you know which is a necessity that that would be fantastic to have something like that.

So I can understand why that might be have an API integration because it might be um you know maybe we're going to be translating stuff on the fly using 11 Labs as a voice uh for responses that we're getting from other tools. Yeah. Well that's it. It's the tooling that's um that's a limitation because yeah I always well I don't always that's an exaggeration but I use um Gemini voice if I've got a I've got a shortcut that if I say ask I think it's ask perplexity actually but the voice mode on that Gemini is particularly good just having a conversation with it it's great for just getting facts and discussing things I remember diagnosing this monitor that I got in front of me because it wasn't working whatever and I was intertwining our debugging session with general chats like normal chats and it was able to just keep interweaving what we were talking about.

And I said, "Okay, so I've started restarted my computer now. What should I do now?" Or, "Does LG have a have a um uh not a returns policy, but like a whatever you call it to get your money back, all that all that kind of stuff. But just having these vocal conversations, but imagine, oh, can you send an email to blah blah blah or can you add a thing to my calendar or do this?" I think the that will be really cool. And again, if you're driving, you're just, you know, cruising with one one hand because I know how you do it. Um, yeah, just uh this natural just understanding and and and my clumsy.

We're all clumsy like when we speak as well. We just sort of don't say sentences when we're not thinking. We don't say sentences. It's clear. And it's great that AI can just interpret that and make it work for us. But yeah. Uh, are you going to shoot now? Yeah, I think so. I think that's it. Right. It's been uh it's it's been a pleasure. Yeah. Um go on to anything.uk. I'm sticking around just for the last few little bits, by the way. So, we're not we're not wrapping up just yet, but uh yeah, go to anything just to make sure that I leave just to make sure that I'm a good lever and it it doesn't mess up your recording.

Do I just click the leave studio button here? Do you know what? I I don't know. But we I can I can remove you and just Oh, all right then. Remove me then. Let's do that. I just want to make sure that I leave my recording behind. Yeah, I know. Because it's got to upload, isn't it? We we normally shut down the stream and then both of us upload. So, I'm going to remove you. You got to say say goodbye and uh I'll remove you and then that should be good then. Well, thanks folks to whoever's been here listening and yeah, check out anything.uk and let me know what you think. Cool. Good to see you again.

Yeah, see you soon. Cheers. Chowo chiao. Cool. Um yeah, so I'm super stoked for 11. Look at me all on my own. This is uh this is uh sad and lonely, isn't it? But um let's jump into I love perplexity. It was massive a little while ago and I didn't understand what people were on about when they when they spoke about perplexity, perplexity this replacing search all the rest of it. Now I'm a pro user and I really like perplexity and it plus it gives me you know um uh what is it called like clock sonnet openai all of the main kind of uh models it gives me access to but what they released this week was perplexity labs and I've got a video coming out soon which I am pretty excited about where I compare a few agents.

Now, I think agents are going to be incredible. Certainly for people like me running a company, but certainly I I recommend you getting involved in is, you know, my quick analogy of agents is if if one person could do one task, that is chat GBT. You you've got one person right in front of you doing the one thing, right? You're copyright. agents. You've got a team of people doing many different things all at once. And you can set them off on a big task to go do their respective things, go create some pictures, go record some voiceovers, go call someone, do this, and bring them all together. Perplexity just added labs to their plan. You can even I'm in Perplexity right now, right?

So, if I just do this, I can set it off. Click this little button here and um this is a labs feature. It will go off. take 10 minutes and it will just do everything for me. And it's really really good. And I've I've set this to like redefine my business. I've I've set up marketing plans and doing things like that. Really really cool feature that they've added as part of their 19 bucks a month. I think this is so Perplexity is a is a great alternative to things like um Manis is uh Gen Spark, which is the two videos or the two agents that I picked them up against. I'd love to know if you have any other agents.

Uh because I again I just want to use them and see what one does best. I don't want to ruin the surprise of my of my video. But the point is Perplexity have introduced labs. So I'm pretty excited about that. This one real quick is just bong. I just want to it's not web focused. I just want to keep tabs on this idea which I've basically described as the singularity. It was started with absolute zero which is a self-learning model or maybe you wouldn't call it a model but it's a self-learning mechanism to for an AI to start with zero data absolute zero and it learns and sets of self tests and judges whether it's and it evolves through that and then like a a week later after that paper was released uh Google announced alpha evolve which is I guess a practical implementation of that self-learning mechanism.

But now we've got sakana.ai and uh you know I'm not I'm not a massive fan of quite a childish looking website but it's still pretty cool that they've then taken an extra level and they are writing its own foundational model here. This is kind of what this this diagram is explaining and why this is important is the GE uh Google evolve and the absolute zero it's sort of like adding layers of its own intelligence on top of itself and it's still achieving roughly the same thing but this one is updating its foundational model and that is I mean again this is this for me is the last remaining um you know uh last remaining like achievement of AI I and and we're I feel like we're just kind of done.

I don't know whether I'm talking crazy talk there, but I feel like we're just kind of done with AI now and it's just about getting um better. But this you stick this in a robot. God damn it. That's uh that's scary talk. But this is great. And it's called the Darwin God machine. And I don't know why, but I'm semif. I'll be honest with you. Semi- freaked out by that name. Join the chat if you are here because I'm on my own now. So you can uh yeah, feel free to ask any questions and I can be more than happy to answer them for you. But um yeah, amazing. And and the where was it about the uh the what's it called?

The benchmarks were just really good benchmarks. I think I got a note down here. Um 20% to 50% uh coding tasks like improvements in 80 iterations. So it's a real like here we go. There you can see it there. Um it's a real um innovation in terms of like it's actually like tangible differences or it's making a notable difference uh in the in the in itself basically. It's not just writing or learning for the sake of it. It's improving itself like self-improving. So this is this is insane. Um so that's it. That's the week's command AI. I've really I've really enjoyed this week. Let me know. I always want feedback with Kabaza hasn't been here.

If you've just joined us, Kabaza isn't here because he's got some he had some um other stuff to attend to, but he will be back next week. We we were Chris joined us. We hopefully all three of us were going to be here, but we are going to be looking for guests. We are going to be looking to bring people on to kind of share on the news and share their perspectives and things like that. So, you got the you got to let me know. You got to let me know if you want to be a part of it. Um, shout out as well to my I really hate.com. Um, I should put that in the banner.

I really hate ai.com, which I'm just determined to I just really want to understand why people dislike AI. Sounds like a meme, but I just I think this is really good for my own personal development and just for me to share with you guys like why there is such resistance because I feel like we are in this bubble and I'm fully in on AI where there are people out there who don't like AI and I want to know if I'm missing something and I want to know if I can communicate or how I can communicate with these people. So yeah, go to I really hate.com if you hate I AI or you just want to leave me a message.

I don't mind if you want to leave me a message on there. I'm not tracking your data. I'm not doing anything. I'm literally just recording your message and uh that is it. Like I can even maybe I can even go into the analytics here. I've forgotten my password. It's uh whatever. But literally just record your message. Anyway, all that to be said, thank you guys for tuning in. Thanks for tuning in to Commander. Join us here every single Well, we try and do every single Friday at 400 p.m. UK time, 5:00 p.m. Central European time. Try and do it live. all there will of course be a roundup going out on the weekend and there'll be shorts and all the rest of it.

But if you do want to join the chat, you do want to join in the banter, then feel free to come join us live. Um, next week we should be back on schedule and of course leave us any find me on Twitter is 0x5M5. Let me know of any really interesting AI news that I can either make videos on or or cover it in the weeks command AI. Cool. Thanks guys. Peace out. Keep on vibing.