#2 Cursor and Windsurf Release! Gemini Implicit Caching & Notion AI Notes

can trigger some of these interactions with JavaScript and and all that lot which I've always wanted from the from the web flow native interactions. Uh so that's kind of cool. Um and I wonder what will happen because you've got Club Greenock, you've got so many plugins and things like that. I wonder how they will be integrated. Uh if they are all free, they are all free. They are all free. But how did they manifest? because you've got draw SVG, you've got physics, you've got so many things like this, whether they will be revealed through the interactions panel or whether they advise that you kind of go into custom code to gain access to those um those plugins.

Do you know what I mean? Yeah. Have you seen how it is right now? Uh what what's it the how the integration is? So if we go to settings, we have this GAP already. Okay. No, I haven't seen this. Yeah. Yeah. And here I can turn on Gap and if I include it, I can be specific about what plugin I need. So I would I would guess this would still remain and what they will do is if you use for example split text in the animation, they will automatically check this for you here. So this is um so it'll be interesting to see if you check something here and then it reveals an interface in the interactions panel or again they say yeah the other way around probably yeah the doing is loading all this is doing is loading the library and registering gap that's all this will be doing so um cool so I vip coded something uh two weeks ago and I was asking AI hey tell me if I need a specific library and I will include it.

And the AI was constantly telling me, well, that plug-in is paid. And I'm like, man, you're you're not updated. It's not paid. It It's a few weeks behind to be fair, but no, it's not paid. And just stop it. I I have access to them all. It was resisting AI. This is free. Let's assume it's free from now onwards. There we go. Don't worry about my wallet. Yeah. Yeah. Yeah. Cool. So, with that being said, very cool. Let's jump into what's up this week. But well, like I say, last week was a lot of releases. Although, to be fair, there is there is a stream right now, an uh OpenAI codeex live stream right now with 10 10k people watching.

So, a few less here. Um, I look forward to catching them up with that. So, uh, open have announcement and we'll we'll we'll maybe tweet about that. But in our things, if you share my screen, if you get my screen up there, um, we'll we'll dive into it. Um, basically, as you as you get that up. So, um, are you sharing my screen? Yeah, we got it. Cool. Okay. Cuz I can you can zoom in a little bit. Feed. Uh, no. Wrong one. So, this one's actually scary. I think I think this is um I think you're sharing your screen, but Oh, no. You've got it up. I'm just watching the stream come through now.

Cool. Sorry about that, guys. So, I actually think this is a lot more scary than ever anyone ever than that's being made out to be. So basically Alibaba um I think I thought it was Tik Tok to be honest for some reason or by dance but Alibaba have released a zero zero uh search or basically a zero data LLM. Um it basically or am I getting confused here? I don't know. We'll keep going with it. But basically it it requires zero data to learn from it. It basically learns itself. It sets itself parameters and then um learns a bit and then like you know documents it its progress and then whether it succeeds or whether it fails and it self builds basically from zero which is just crazy which is a wild thing to think about because it you can just set it up.

I'm sure it's not as simple as that, but the cost savings are amazing. They've they've they've mentioned here it's 88% less cost reduction mechanism. Um, which is quite interesting. I was just looking at this now. Um, 64,000 search queries using Google would cost around $586, right? Whereas um a 14 billion parameter simulation on a 100 GPUs would only cost 70 pound $70 and 80 cents. So there's efficiency gains here as well, but it's just Yeah, I actually think this is um something slightly different, but this is still this is still quite crazy. Um there is um what else we got here? So, I actually think this is sing a singularity. I think that's this is what they call singularity, which is like I I I released a post on LinkedIn and put a photo of the Terminator up.

Yeah, it's really getting difficult to get like um get to, you know, being able to predict what is going to happen next. And the rate that these companies are moving right now is just crazy. like the rate that they are uh putting out code and putting out uh products is just going through the roof. Just reading um the first few lines of this article, it seems like what Alibaba has done is allowing LLMs to not use search engines and still have search capabilities. Meaning that this is like double bad for Google and Microsoft. Like not that humans are not going to use their search engines, LLMs are not going to use them either. So this is like crazy scary and we already see by some estimates like 70% reduction in search.

I was chatting with a client last week and he asked me he he wanted to do a campaign and then he asked me a question and as he asked me he started asking chat GPT as well and the question was more like Google oriented and I was like you just answered yourself you yourself not a tech guy you didn't search it in Google you asked Chad GPT for it and I think this is like this this point that everybody should realize that like search engines are pretty much as as good as dead. I don't know how it is for you but if I Google something I see this Google summary like AI generated summary and 90% of the times maybe even more I'm good with it.

I don't I don't read even the first link. I don't click anything. No. No. For sure. It's funny you say this because I sent out a tweet last week. I don't want to dox myself or anything now, but uh I sent out a tweet this morning and I I had Samuel here um arguing with me. Very polite. It was all good, you know. It wasn't like a full-blown argument, but my my uh my three my tweet uh yesterday was SEO is dead, I reckon. So, it's so funny that you say this. I think we still got a bit of a way. I think SEO is absolutely still relevant and Samuel makes some good points here like but I do think we're in because I think people's the way that people are searching is very different now.

Yeah. Um I just think it's Yeah, I don't know. It's I agree with you. Anyway, I want to I want to uh announce that I made a boo boo. So, basically it I'm getting two articles mixed up here. This is the one that actually is freaky. So, can I move this into Sure. Um, so this is absolute zero. All right, I'll wait for this to I'm hoping this will share. Um, oh, holy Zero training data. This is zero training data. Self-evolving mechanisms is what we're talking about. Let me make that bigger for you guys. Right. And this is what I meant to say about Terminator. the fact that it has no it it has no data and no data sets uh no human labeled examples to guide it just a system that evolves and improves entirely on its own.

Um I have a feeling this might be bite dance which is why I was getting absolute zero reasoner. This sounds like uh awful to a philosopher zero reasoner but interesting. Yeah. Yeah. LM without data. Isaac Newton born with teeth. That's what S say uh let's go through these t TLDDRs. Absolute reason model includes uh introduces self-evolving AI allowing autonomous learning problem solving without relying on external data or human supervision marking paradigm shift in AI AI development. I think this more people need to be talking about this for sure. reinforcement uh learning with variable rewards. What that means it's um it's giving its own it's giving itself um the the the parameters in which it knows it solved the problem.

Whereas right now I think there's three different ways we we train AI. But right now the the human is the intervention of did the did the machine get this right? But this one it tells itself that it's right. basically uh a absolute zero demonstrates cross domain generalization ex excelling in diverse fields as mathematics and coding. So you know arguably some of the biggest things you know uh at the moment are probably I would say physics is probably only um bigger than that. But how can it how how it how is it that it doesn't do any reasoning but it can do mathematics because isn't it mathematics just reasoning? Sure. I mean it basically I mean you you talk about like humans um we evolve through just well this is how we evolve.

We go from unreasoning um little kids who just don't don't know how to take care of themselves and we learn through baby steps quite literally um just to get to a point where we're able to solve math problems. You know, I think it's I think we're probably underestimating the time it takes to actually get to a um mathematical problem. This is really interesting because super interesting. It is and that's why I'm trying like to understand here because we have reasoning and intuition and we shouldn't like mix them to you know uh intuition maybe you build intuition with reasoning or without it but like you build intuition to with having experiencing the world as a child or as reading something maybe you build intuition but reasoning is quite literally tapping into the logics and going the actual steps of maybe even the the scientific method to to reason and to to build up you know a a situation where you can replicate it.

But as of now AI that the reason why they hallucinate is because they try to mimic our intuition. Okay? Because if you so if if you tell an AI if you tell an AI what is like this huge number multiplied by this huge huge number is they give you a very interesting answer. The answer is something that the numbers at the start are correct. the numbers at the end are correct but the middle is not correct because the AI has an has developed a sort of intuition for what is the multiplication of these two numbers but doesn't quite do the actual reasoning and m multiplication of the numbers. So it's roughly right but not quite.

And that's that's why they are so good at guessing the next thing but they are not good at like being 100% sure with the next thing. It's just intuition. There's also Yeah. But there's also this thing of something like lost in the middle or something we need to be aware of that they're getting better at. This is this is more you know to do with coding and stuff is that you you build a context so big that the middle bit just gets a little bit jumbled but what you're is what you're saying something something slightly different. Um no I think it in a way it is related but it's not about the amount it's not about the context window.

Mhm. It's just reasoning is um based on like hard data. Let's let's let's say in multiplicating uh two numbers it's like the actual like the exact number but intuition is when I say what is uh you know 100 times 100 you know it's more than 500 you might not know the exact answer but you know oh for sure it's more than 500 because you have that level of intuition now AI is very like getting really good at getting super close but it's not always exactly there. That's why it if you ask in programming, if you ask it to to write something, it sometimes it comes up with an API that doesn't exist. But it would make sense for that tool to have an API with those like methods.

They don't have it though. The AI just intuitively think there that should that thing should exist. So it makes it up. So is that's making int making intuition up basically. Yeah. Yeah. Wow. Yeah. This to me absolute zero is I think a turning point in AI. I don't I mean I don't know whether they've actually like this is something that's like live or whether it's like a a paper or of some description which that you know that it's not um it's a theory basically. I'm not I'm not sure how this has manifested itself but manifested itself. I don't know. Um, but either way, this to me was just like, "Holy crap. Holy crap. We're getting there."

Self self-evolving me. Yeah. We're getting where we're getting where. That's the worry. Yeah. Event horizon is just there. Yeah. Yeah. Exactly. Yeah. Um, so yeah, that was uh pretty good. Let's um start going let's bring it back to web because that's scares me being uh being there. Um so implicit caching. This is nice. This is happy stuff. Uh Google adds implicit caching to Gemini 2.5. This is this is actually really really helpful and I think I wanted to raise this because there are a few things that you can do to actually take advantage of this. I love you know what caching is, right? I mean, yeah. But what is implicit caching? It well I mean it's imply So basically it it um the AI will detect whether you are actually using the same words, the same phrases, whether it's got it and it will it will cache it yourself itself.

Right now you have to actually um try you have to actually do something to make the make it cash like you need to actually trigger a caching. This is if if if you use similar and this will get to the crux of it really if you use similar words it will just cash and automatically cash and automatically like save you money at the end of the day. So I think this is good because we can take advantage of this and that I'll get to the sort of um quotequote downsizes at the it needs a minimum uh request size which is,024 tokens for 2.0 and 2.5 is48 tokens. This is really easy to achieve. It's not like this isn't like you need a whole code base to actually you know trigger this implicit caching.

But what what they're saying is that if you start it might kind of be relevant to what you were just saying now. If you start with um common prefixes, which could be I was thinking about this just before this um this call here is that like it could be as simple as saying you are a software engineer, you know, or just you having similar prompts to trigger that caching. Um, and this is so valuable like at the end of the day, especially if you're being charged for um for for each like request or whatever going out. This is actually really cool. Is it like system prompts in a way? Well, kind of. You could probably use a system prompt to trigger it.

I think the um cursor and wind surf both have kind of um you can have your global I've got a video coming up about this actually. You have your global rules that you set on every single codebase, every single project blah blah blah, but you can actually have project rules which you can include in your prompt. So that might be a nice way to do you know to to trigger the implicit caching. And I I really like 2.5, although we'll get into some new releases. Of course, there's some new releases, but um I do like 2.5. So, um I think that might be a I think a lot of people do. A lot of people are really happy with 2.5.

So, I think it's uh if yeah, that'll be a really clever way to always trigger this implicit caching, which you should get. Saying that, I think they just benefit from the speed because I don't think the updates that cursor or wind surf have made actually they're trying to simplify their pricing. And in that I think a lot of them are doing it per request as opposed to like how many tokens you're using. I think they're just taking the hit on small tokens, lots of tokens or whatever. Um so I'm not sure whether we actually benefit inside of the coding tools, but we certainly could benefit inside of let's say Google Gemini or something like that.

So, um, this is a really nice little feature that's been added, I think, and I think that, um, it's going to it's just going to help reduce the cost cuz I think it's growing. I think I don't know about you, but I'm spending not not a lot of money, but quite a bit of money on my um, AI output, but I'm trying to do it as cheaply as possible by actually using the APIs as opposed to the kind of chat platforms that are out there, you know. So, um, yeah, bit of good news there. Yeah, that's great. The the cost is definitely growing, so that is welcome for sure. Uh, let me just get my notion notes up here because I uh I want to go in a certain order.

Um, okay. This is this is my one of my articles that I wrote actually. I do post on LinkedIn to be honest. This is actually this I really welcome your thoughts here actually on this one. tonight because um Cler kind of announced that they were they went all in on AI. They went allin AI strategy and I think they fired a lot of their workforce on um because of this AI strategy and this gets into a quite a big debate to be honest. Um, and they've now basically backtracked on their decision on this because they've just started to receive really low quality, you know, results. They they they replaced their um customer service as well.

Um, and just well, you can see here quality and empathy were sacrificed. Mhm. And they're backtracking on it. So, I mean, it's good to see that they've actually, you know, recognized that going all in on AI, it was a mistake. Um, but at the same time, you know, very, very recently, Duolingo declared themselves AI first. And I think they're going to realize the same thing. And I think I I saw a video recently from a guy, YouTuber in London, actually, is a US guy, um, noticing the deterioration in quality of Dualingo. Do you use Duallingingo? I don't. Um I don't. Yeah. I learn the traditional way through hard work and craft like the good old days.

Not like a my English is not AI generated. I don't do skiibbidity skiibbidy. Um but he was saying that the stories which uh in Dualingo were particularly kind of just sucking basically. So, and we'll get into a discussion. Did I mean I don't want to skip too far ahead, but did you watch the video that I I put in the notes for the for this? No. No, I haven't actually. Okay. But anyway, this whole idea of like um replacing people with AI and it's a conversation I've had with a few people this week actually. you know, this idea of like entrepreneurs being empowered to build their own tools to, you know, I'm replacing massive teams um with AI.

Like I'm getting so much work done because I've got AI. Granted, I'm not multi-million dollar company that is able to afford people like I I I feel like this is empowering people like me not to um not to justify my own use of AI. I can't afford to hire people. But saying that, you know, you're empowering all these people to to sack off people and fire people. I've seen I'm seeing a lot of um redundancy. Uh actually that that's actually really relevant. Um was it Google or Facebook or someone massive just fired so much of their workforce? I don't know if it's to do with AI, but if you're hiring people, uh, sorry, if you're firing people because of AI, then no one's got any money to, well, people are not earning any money.

They're not they're not able to there's not disposable income for them to spend the money. And so like we're going I think if we're not careful, we're going down a bit of a rabbit hole of total annihilation of ourselves because no one needs to work. So that no one's got there's no reason to start a company because no one's buying anything because I don't have to work because I do you see what I'm saying? It's this weird cyclical thing. But that is far in the future. I don't honestly see this happening anytime soon. And the reason for that being Open AI, did they buy Windsorf or are they still in talks? I heard they they um no I I'm not going to say I'm not going to say I heard an update that they are they've committed to that three billion.

So that that's the crazy thing that that's like my take on it and it's also relevant to Dualingo calling themselves AI first or CLA you know going back to the old good old days. I don't think these AI tools are quite there yet to take any job. We thought something like customer service would be gone by now. Like if you look at chat GPT 3.5 like the earliest versions that we all saw and we were like holy like this is the new era. Now you do not use that version. You you find it to be like awfully bad, right? But we saw it and we're like this is going to replace customer service for example and it's still not there and we see clar now going back to humans and when it comes to coding I think that it's even like further away in the future if it happens like ever.

I don't know. I'm not going to say it's not going to happen ever. But it's far in the future. And a good reason to like to think this way. At least that's that's how I think how I came to this conclusion. Why would why would open AI pay $3 billion dollars for a fork of VS Code? VS Code is free. It's it's an opensource code editor with a splash of AI on top of it. Why are they paying so much money? Windsorf doesn't have users. So they're not paying for you know getting access to those users. They don't also have like decades of work on that uh tool like windsurf building windsurf didn't take decades.

It's not like so much code that is so difficult to replicate or to write again. I mean cursor and windsurf they are going neck neck to neck right they are like very close. So I I assume the code is not in itself is not like too difficult but at the same time difficult enough that open AI would pay $3 billion for it and they are on the top. They have all the money from you know Microsoft and Satia. So why don't they just create an internal team and fork VS code? They don't do it because probably it will take time. it it it will take a lot of like engineers. So it's still like they can't tell their own chat GPGs and press a button and say yeah create a fork of VS Code.

Plus they need to compete. They you know if they launch their own that's just adding to the to the mixing pot of brow uh code editors then. So I just don't think they wanted to compete. It's like we'll just buy it and then you know Yeah. But why do they pay three billion? like I if if if if it's so easy to create software so that companies are already firing um you know programmers then why shouldn't they you know fire programmers and just use AI to create windsurf in in opposed to paying three billions for it that's what I'm saying I'm saying that that the software is still valued the only thing that has changed with AI the way I see it is like difficult program problems are not difficult anymore.

Things that were like almost like impossible to do or not really wise to tackle. Now you can say like yeah let let's take a few like good programmers and tackle that impossible thing because we have AI we can we can move faster but it doesn't mean that uh programmers are obsolete. Yeah, that would be my take. Well, I mean, well, you can see on the screen that they I mean, reaches an agreement to buy. So, I don't I I'm not sure if this is um what we were speaking about last week or you were speaking about last week, but I mean, there's your answer. Um but yeah, yeah, by the way, on why they are paying three billion to to buy this, um Primagen has a really good take as well.

He mentions that probably because creating a soft, you know, a piece of software is not a a piece of cake as like some of these companies try to sell the idea to you to um but also because open AI what is what they are lacking is the integration of the code in real world project they they they they kind of get they know what code output they are giving giving to you but they don't know how you are using it. So by owning this tool windsurf in this case they will they know what they are giving to you but how you a developer is how you are using this code in your codebase they get the full context.

So it's about data. Yeah. Well that's it isn't it? So it's um there's no surprise that they've well yeah that's what that's what worth3 billion dollars apparently. Um, funny enough, I just watched a video just before this call actually and I, it was by a guy, really good video, um, talking about how vibe coding has been sold, this this lie of vibe coding has been sold to be I mean, it always happens. It happened with Facebook, you know, uh, Zuckerberg took a shot in the dark and it was like obviously blew up and now he's a billionaire. Vibe coding has kind of gone down the same route where people think, "Oh, cool. I can just build a tool now."

And I'm in conversations with someone at the moment who wants to go down this vibe who wants to leverage this vibe coding um you know movement and I'm like I was saying to him literally this week because we have call every week you know I think you just need to get your app out because you know you you build it you spend all these time building it you get to MVP because then the moment you release it is the moment you realize that no one wants it and then you go off and build your next tool or you go you know it's such a small task in the grand scheme of things to build these tools because then one, yeah, you've got to actually get people to buy it.

Two, you've got to maintain it. You've got to like fix bugs and develop features and things like that. Like you, it's not just like you throw it out there. You are you are literally committing the moment you release it is the moment you commit to potentially the rest of your life. You could be for the rest of your life supporting that application. And I think people don't realize that it's a full-blown company. It's hiring people. It's it's churn. It's managing churn and all this sort of stuff. And yeah, I like I said, I just watched the video just before now. Um I don't think people are realizing that there is again with with open AI, you know, building their own thing that it's it's a lot more work than you think.

Yeah. You know, so much more work than you think. Um probably well costly as well. and you might it might not succeed again. If they were to release their own one, it might not it might not compete. It might not be as good. Um but yeah, it's uh it's a really really interesting kind of state of work at the moment is kind of yeah really just all up in the air to be honest. Um I see you've got um Upwork on your screen. Are you I'm not sharing it actively right now with um but now now we are sharing. Yeah. So I don't know if it's the article or or not but somebody mentioned it in today's call u that Upworks uh co this is probably not this is the fall release 2024 but upwork co so I'm just going to paraphrase it mentioned that kind of like not too harsh but somehow in a harsh way to its own staff that they that you need to be prepared for the age of AI that we expect you to basically utilize AI which is not a surprise but I think this is what a lot of developers are not realizing for the entire life today I've been like on the side of developers but for a moment and to be fully realistic it's a lot of people are not uh you know utilizing AI and I see this issue not just with developers with pretty much everybody uh graphic designers anybody like with AI you can just get so much done so so much faster and a lot of people are not using it.

So in the market um you know obviously time matters and you know the the next the next founder who thinks about hiring you might contact you and ask you how long does it take for you to do X and Y Z and if you say like 2 weeks and somebody else says well I use AI and it will take only two days you might not get the gig and you shouldn't be surprised. So that's how your job will be taken away. That that's essentially how it is. Yeah. Yeah. Yeah. Totally. Um you can fight, but the more you fight, the for the more you'll falling behind. And it's not it it's not to say that AI is going to take over.

It's just going to be a a tool that you can leverage just like computers are a tool that you can leverage. You know, it's it's it's an inevitability. And I think that um developers are pretty cool. I think a lot every pretty much everyone I know is using AI. I don't think they necessarily need to go down that vibe coding route. Uh vibe coding is obviously this um debatable topic certainly at the moment but absolutely at least at least use you know cocomp completion and things like that and if anything to to sort help use chats to generate the functions that you need rather than writing them yourself you know. So, but yeah. Um, should we move on?

Yeah, sure. Any more thoughts? No. It's like this is a massive topic though, the workforce. Do you know what I mean? Yeah, it's I've had too many of these conversations and people are so worried that that's why I'm also trying to cover it every time because people constantly ask me people the the the worst thing that this AI is doing like it's paral what's it called what is the word paralyzing parallel when when somebody cannot um move what is the word for it yeah paralyzing yeah paralyzing it I this is what I see AI is paralyzing ing so many people because they are afraid of going to any field at all. They say if I go into this AI will take my job tomorrow.

If I go into that, AI already has taken over. I'm like no, it's just so far from truth. And the difficult part is to understand is um being a junior to in today's era, it's very difficult if you don't use AI. So you cannot you cannot get a job as a junior anymore. Not at least in the traditional sense. You can be a junior and get a job or like an internship as a junior AI vibe coder. I could totally see that. But if you are a junior in a traditional way and you're like I have no idea and I don't know even how to use AI then I'm sorry like this is this is the world we are in and you have no place in like like in no workforce but maybe the answer is to just like how it was in the past pick a subject that you are really interested in and use AI to its full potential.

learn to use AI for that subject without thinking that AI is going to take over your job. No, think that with that AI and your interest for that subject, you can build something potentially even like better or like at least um more impressive in the first site better than a a senior developer. you can't do this as a junior and I' I've done this myself and you know my senior dev looks at my code and he's like this is dog and he's right but uh but he also says that he couldn't do this in a day what I do in a day he couldn't do it so the code is do yeah I agree and I'm never saying like it's not but at least I can get the idea out I can test it I can get it in front of people and once it's time to rewrite write the code or whatever.

I can either get better with AI and do it myself or give it to a developer who can use you know his his knowledge plus AI to rewrite the entire codebase. Yeah. Yeah. And that's the important thing is like as with everything that you got agile's way agile way of working which is getting the idea out there because again going back to the conversation I had with my friend you you'll learn that no one uses it or they use it in a completely different way than you expect it or the the idea is completely different. I went to fig conf or config rather this week represent and um would you believe oh what was he the the founder said it was a it was a gif generator.

Figma started out as a gift generator. Wow. And obviously now it's Figma, you know, like you it will never be thing that you you launch and if it does even if it does get successful, you know, you it's a shot in the dark. So just get your idea out there or not in in your case, get your code up there so it works. move on with your life because I don't know about you, but once I get something up or I get something built, whether it's the best code or whatever, there's just a very a massive like um weight off my shoulders, a relief in my mind that is allows me to think of other things and then, you know, I will return to that.

I will think of a better way to do it. But um I think this is a really nice segue into the next uh not even article but if you share my screen there there was a um an emergency debate that was brought together by Steven Bartlett and it was sh like um the founder of Replet and I was trying to find out who the other guys were but one of you got one person who's kind of in the middle but you got the founder of Replet who um no offense to the guy, but he came in looking like a super villain with the ginger goatee and the red jacket, but he was a very smart man.

And um another guy who, if I skim forward, this guy um let's have a look see what their names are, but he was against um are you going to skim forward? This guy, he was against AI and he was he was he was very fearful of AI and how Yeah. work forces are being sort of replaced and um we sort of missed the boat a little. He wasn't a doomer like he he he was very there were certain things that he was very excited about but he was very much like providing that that other side other you know that consideration of why we should be prepare I I want just yeah I guess preparing ourselves for uh the inevitability that um we're self-destructive with with the onslaught of AI and whatever and there'll be always people trying to capitalize on, you know, there'll be plenty of people doing good, but there will be people who are right now, right in this moment, building things to um take advantage of people.

You know, the easiest one being um uh imitating voices and then calling you up, pretending you're you're granny and I need some money and can you send it to me? Scams, right? So there are people already uh doing terrible malicious things with AI, but you know uh I'm not going to like I think this is certainly a watch because I'm trying to understand why people just the moment they hear it's AI, they just like nope, don't want to read it. Nope, don't want to watch it. Funny enough, half the stuff they're reading now is AI. So half the time they can't even tell it is overwhelming. That that is why, isn't it? Well, I mean, I'm not talking about whether it's overwhelm.

I think that, you know, I listen to a lot of um philosophical and self-help videos on YouTube, particularly if I can't sleep. And I'm sorry, I know full well that the the audio is AI generated. I know full well the script is AI generated. I know full well the imagery that I'm looking at, if I'm watching it on the screen, is AI generated. I know this. But the bigger picture to me is that it's helping me with my mental state or it's helping me um solve problems and and just make sense of things which I haven't done before. And I know it's AI generated, but saying that it's not about me. I I flip down to the comments and I'll see people and they'll be like, "Oh, this is AI junk."

And whatever. And it's like you you're just you're automatically just throwing it away because you know that it's AI. Whereas for me that's actually finding this really helpful. I don't care that it's AI but people just switch off just because it's AI generated and they AI slop you know AI oh this is AI slop and it's like just you have this you have it's a you problem you know I don't know why people switch off but I want to understand why people just switch off just because it's AI. Well, the the reason for that being uh I was dis uh discussing this actually with my philosopher friend uh just yesterday and I think this has a lot to do with how we find value in things like if something is hard to do and takes a lot of time and basically we can assign effort to it we automatically think it's more valuable.

It's not always onetoone like this. It's not always, you know, more effort means like, you know, more value. But it is in many cases like a Ferrari, you know, valued a lot because the, you know, the amount that they output is not too a lot, you know, they cannot make too many of them or, you know, I don't know like any of these super cars. But it's also because really it's really difficult to make. I know there is I I mentioned Ferrari on purpose because this is a luxury brand. So there is this element of luxury that we also assign value to it. But anything that is very difficult, we again assign value to it.

an AI slop. The issue with it is because that that's that's how I see it is because you don't you don't see any effort being put in towards it. Even though sometimes the the output as you said it can be really high quality. I'm not I'm not saying it's high quality. I think the bigger there's a bigger picture here like it's still it's still crappy. It looks crappy or whatever. But um but yeah ultimately I I am trying to understand the other side of it and I am trying to understand the the like they discuss in this video here like the the sort of downsides to it because I don't want to be just blindly running into it.

I want to try and meet in the middle somewhere um and know that I'm doing using it for good and providing value rather than just throwing out AI slop that's not um you know not relevant to anyone. Um and you said something earlier that was like quite relevant in that people we talk about AI taking over people's jobs, right? It's not taking I think that word take over is an interesting one because it's not it's not about taking over like I want to encourage people to do use I think there's this perception that you need it to do everything and we'll maybe get into AI agents a little bit but I've been looking into AI agents and AI super agents right it's not about it replacing an entire workforce or it's not about replacing your entire role I think we need to look at it in a more modular way and that it's empowering you to do certain aspects of your work.

For me, you know, I take these talks that we do, I take phone calls that I I I have. I take videos that I make, I transcribe them with AI. I turn them into blog posts with AI. I do little bits and bobs. Yesterday, um I had a call to do to discuss a contract. Now, we haven't sent this contract out to the client yet. It's an internal discussion and I uploaded the original contract that I had drafted. I up I uploaded the transcript of the discussion that we had about the contract and I said everything that was mentioned in here where is it mentioned in the contract and if not where should I put it?

And I was able just to like cross reference that then and just use AI to help me make sure I'm covering everything in the new contract that we've written. And I just think that's faking amazing. I think that, you know, and that's just a small aspect. I'm not replacing anyone's job. I'm I'm enabling my own job. I'm enabling my own work through the use of a AI and little bits and bobs. Um, but yeah, it's it's it's interesting to see how the other side perceive AI and and um yeah, also the downsides. I I do the same even with my team. I don't think that, you know, any piece of AI coming out, I don't think it's going to replace any of my teammates.

If anything, I want to get my teammate get their hand on that piece of AI so they that they can output more. That's that's what I want. I don't want to, you know, myself go to the AI and I don't know like start prompting or whatever or play what around with it. I want to know okay there is this AI that can do X and Y and I have this person who is uh very good at doing it as well but with AI they can do 10 times more uh work with a higher quality that that's basically what I'm looking for and I want to give their you know give the AI to them and when it comes to hiring this is where it becomes difficult this is where AI is not taking over your job but AI is making it very difficult to hire you because if you're not using AI, I don't know how I can hire you.

Yeah. Yeah. Um let's move on. I don't want to keep doom and gloom. Did you want Is there anything you wanted to like bring up at the moment because I'm going to move on to AI uh notion? Yeah, let's go. Yeah, this AI notion notes. It's like uh Granola is in danger. Uh, I have uh I have just I've got really nerdy about this. I don't know why. Maybe I maybe this is just me that's super interested in this sort of stuff, but um I'm really like okay, so let's let's let's backtrack, right? I'm dyslexic. I think I mentioned last week I'm dyslexic and having conversations um even now even this conversation absorbing the information in real time and being able to process that and and very clearly uh articulate myself is a massive challenge.

I had a phone call yesterday that I was like like I I try and avoid phone calls especially with clients because I just I'm having to think and understand things and it it I I know that I come away and I think a crap I've forgotten something or I didn't have a pen and paper or you know I'm I'm worried that I might have missed something. So when AI comes around, I I was straight on it with AI notetakers and actually I had a podcast called that tech show which we used it so that we're able to actually just summarize and write the description of the podcast rather than actually you know any sort of notes.

I've been using Fathom for absolutely ages. Fathom is um maybe I can get it up here. Um, Fathom. I've been using Fathom and the uh this the notes in this app were absolutely incredible. They give you action items. They give you what the call was about. They just and you're able to chat with the with the transcript as well. Really, really cool. Um and then you and I obviously did this thing where um well actually just to round off the dyslexia thing, it was just amazing that I'm able to uh go back and and watch things and understand things in a different way. And also the AI will infer things and pick up things and make connections with things that I didn't actually make that connection like oh the client repeatedly mentioned this this and this or whatever and stuff like that.

Um, so yeah, then we got this deal with Lemy's newsletter, Lenny's newsletter or whatever, and they um they announced Granola, which is I don't know if I can Well, I mean, it's not up on my screen anyway, but maybe I can just do the uh Granola. Granola is really cool because it runs on your machine and it takes the input from your microphone and the and the you know I don't know how it does it but the input from the other person cuz Fathom joins your call which is which is you know it's kind of intrusive but it's it does the job. Um and you know if your internet drops it's still picking up the call of the other person whereas uh Granola it's all on your machine.

Um, and what's nice is that you can actually take notes. As you can see here, you can kind of take notes as you're in the call as well, which you can't really do and fathom. Really, really liked it. Used it for a couple of calls. Could see the live transcription. And then notion drops the bomb that they are doing the same thing. And I liked granola because it actually felt like notion and it would copy directly into notion work perfectly. But now notion have launched their own um app, meeting notes app. And it is pretty good like most most businesses I know I say businesses you know most most people in our sort of level of business notion runs their company.

I mean do you use notion? I mean you use notion but how how deeply ingrained is notion in your so everything from finances to hiring to team management to documentation literally all my life is on notion. Christmas gifts, notion. I have everything on notion. So, yeah, I don't use their mail. I don't use the AI yet, but I have been a Notion pro user since they were in beta. I saw a post just now. Everything just happened just before this call. Um um Chat GBT is powering AI in notion. So, I wonder if you're actually just getting chap GPT in um in in in the AI for notion. Regardless, um yeah, my all my documentation is in for my business is in notion.

I want to use it more. If there's one criticism I have is that it is it's so powerful, but it can be very like quite messy to like try and organize stuff. I think that's what you naturally get the more powerful a tool gets. it just gets a little bit convoluted and and I probably need to spend more time in notion to get hold of it. Saying that with these new AI, you know, um AI notes equally, it was a bit tricky to set up and the integration isn't as nice like um all of like Fathom just pops up into the right hand side. I can start the call with Fathom. It just goes into it.

Whereas I need to now create a new meeting or seemingly I don't know. I may need to look into it a bit more. create a new meeting um note and set off the recorder. And it's doing it. It's recording this call right now. The notes are great. The action items are great. I I'm not sure if they're as good as uh Fathom if I'm honest. I still need to do a few more calls, but they're they're they're pretty good. I mean, I've got I don't know what the whether this is included in that Lenny's newsletter because I have uh Notion Pro now. I would like to try and move all my stuff over to notion, but um I see you've got the prices on your screen right now.

I think Notion AI is different. So, okay. So, yeah, but the the Well, let's go into this. Um yeah, I'm now sharing my screen and this seems to be very interesting. AI meeting notes perfect meeting memory in notion. So just like granola and what's really cool you can transcript you can get a summary you can get a note and this is so good man because I have a weekly meeting with my team uh talking about like how the everybody's weeks uh week was how is it going to be what they will be working on and I have somebody who writes meeting notes still in 2025 can you believe this. I don't believe it. Yeah. What?

Uh because I don't know, man. I I I try to use granola. I try to use a It's just so messy at the end. I feel like what I have because the person who takes the notes is the project manager and he later on assigns like actual tasks in a database in a notion database to to people. So that's why I mean these AI tools I think they are great for summary and note takingaking to get back to them and to get back to the actual you know note and see what was said but I don't know if the actions are thorough and like good enough for me to you know to let them take over and you know do the action because we don't do we don't store the tasks obviously Obviously in the meeting uh note we have the task and you know each project gets its own task and its own database.

So it's very you know organized but also disconnected from this. It might get there one day. You know you mention the the project we mention the project names for each client. We have we use like you know a few letters to mention the project you know we shorten the names but this looks very good and all I see is these companies like attacking each other like more frequent and more like faster than ever before because you know I was about to buy the AI from uh Loom because I use Loom day in day out. It's so good that this is like the primary way I talk to my um teammates because it's async and faster, right?

Mhm. And I was thinking about buying the AI, but now this is kind of like lives on a different tool, but I guess I mean realistically you you could link to this. You say here's the here's the record. You could create a a notion art like a page and then send them the page with the Loom embedded or something. Maybe a bit more hard work, but it just depends on how you use Loom and what the AI brings to it. I mean, yeah. I don't know. Yeah. But interesting this like I'm a big Notion fan. I know they have like their own issues. They have some quite quite some haters, but um I I do like Notion.

Yeah. Yeah. That's pretty cool. Um, I'm still I'm still bit it's a bit cumbersome for me, but then I'm not I'm not living and breathing notion and whatever. I just I just want to start a meeting and start a you know recording it straight away rather than this cumbersome thing. But um in an attempt to keep everything in notion, it might just be worth the double kind of double action of doing it. But yeah, pretty cool. I'm using it and it's free at the moment. I think they'll probably charge for it, but I think for me it's free at the moment. So, we'll see. Well, there you go. There's some price in there. It's crazy.

Mhm. Mhm. Calculate savings rather than actually Clever. Clever. Um, right. Well, let's uh let's move on. Um, I'm two big releases from Wind Surf and Cursor this week. Yep. Um, so Wind Surf have released wave 9, which wave eight was like last week. We just spoke about wave eight, so I'm not really sure what their release cycle is, but it goes back to what you were saying earlier about how quickly these companies are moving. Wave 9 introduced the uh Wsw SWE models that basically I think that they are um they are uh anthropic claude and pretty much clawed um apparently outperforming claude which is pretty um hard to believe but you know there we go approximately claude 3.5 sonnet levels tool call reasoning while being cheaper to serve.

So the SE1 for me is the and they they attempt to address three different levels of coding right uh whether it's a big sweeping change across the code base with complex you know complex uh reasoning and things like that you're probably looking at something like uh uh number one uh one light uh smaller model cascade better quality probably for lighter tasks and uh and things like that and then the mini is good for the tab, you know, in page corrections and things like that. So, uh, yeah, I've started using this today. I was giving it giving it a whirl. Um, seems to be, you know, good enough. Um, it's interesting how because I think they had their own model.

Well, Cascade Base, you can see it here. I think it's just like an an evolution of the cascade base, but pretty cool to have um their own models and they this is potentially why they're worth three billion. Is there any reason for me to start with windsurf? Is it like easy for me to basically you know open the same code base instead of like with cursor just open it with windsurf? Yeah. Yeah. Yeah. That's like the churn, man. The churn is so easy on these companies that that's why they shouldn't be, you know, worth three billion maybe. The the only difference is I mean you've got global rules which both of them respect, but the um project rules are slightly different.

You've got a cursor file or you've got a wind surf file for the project rules. But outside of that, yeah, it's there's no difference. So you can just download it, use it, um, and go with it. So that's pretty cool. I I still have cursor because of that good deal, but I'm thinking maybe if I get stuck or like when I feel like the code quality is not good enough. Maybe I I don't know. I try a different tool and let's see what it does. We We're splitting hairs at this point, the way I feel. But I think that's that's a conversation for another day. Like interesting. Yeah, they're all they're all pretty good. I like that it's integrated.

they got their own models whereas obviously cursor don't they're leveraging other models but then they're free to make um better software which in fact that goes to if I maybe open up the 050 launch wherever it is let me get that up doing some live googling doxing myself or revealing some uh poor search history here we go Uh, this is them talking about uh the pre-release. Yeah, it is, isn't it? Um, I want to see I mean, I can summarize it. Anyway, they're making some interesting updates to the UI and the usability of the the software. Um, I actually really want to get the simplified pricing. This is I need to understand the simplified pricing, so I'm not going to kind of go into it.

Um it seemed like they were charging per request anyway, but then this one's also announcing uh they're charging per request. I even went back to the way back machine to have a look if there's any difference from today than there is yesterday, but the the pricing is the same. It's so you get 500 credits um and a request is a credit, but then you uh max mode, which is like heavy duty stuff, which I should take advantage of really. It's like a 20% fee on top of the the um the request itself. I'm not too sure. So, um, unified requestbased token billing. Can't really speak to more than that. Uh, new tad, um, model, which is cross file intelligence refactoring.

So, this this seems to be popping up so much in people's, uh, videos and things like that. The actual tab, the next to go if you make a change and it knows that you want to make that, you need to make that change in the rest of the file, you just tab tab tab tab tabs. Now it can do across the the files which is pretty cool. Background agent is also really cool with parallel task uh processing. Set set your agent off, open up another window, do something else. Love that. That's really cool. What do you do while your wind surf or whatever uh AI is thinking to or like generating code? What do you do in that 30 seconds of I know time that you want?

Twiddle my thumbs. Take a sip of my coffee. What do you think I'll do? Enjoying life. Gosh, enjoying life. Touch grass. I love that. I I I will say like the the projects I'm doing right now, the level of kind of code like I see people or hear about people setting off 10-minute kind of like requests and things like that. I'm I'm not that brave. I'm doing like little bob bits and bobs where maybe it will take five minutes like at most. So, you know, um I'm I I check the files as it's making the changes just to kind of see how it's doing it. Sometimes it backtracks on them, whatever. Um I'm not too sure, but um yeah, it's uh yeah, it's it's just time for me to touch grass and just live life.

What do you do? Same. I go for a run. Go for a run. Wow. Making big changes. I let the entire code to be regenerated again just because there I wanted to change a background color. Yeah. Right. Let's uh let's let's wrap up. Right. So yeah. Um uh multiple. So this was interesting. This is something I've been triing. I have the back end and the front end in my workspace and I'm wondering I was triing if it would would changes I would make or want to make would be able to just navigate across the whole codebase but it doesn't seemingly now it does multi-root workspaces so you can yeah have your back end and your front end as separate kind of you know running I don't know whether that that computes with you but like you have code for your front end you have code for your back end but they're separate code bases they're not this sorry I need to be on my microphone they're not um yeah it's not a monor repo we call it when the back end and the front end are all in the same code.

So you don't do that having all the code in one. There's a time and a place. I'm not going to criticize there's a time and a place but um you know I deploy on separate servers. So when they're deployed on separate servers they No, that makes sense. Yeah. Yeah. Yeah. So uh long file optimation, enhanced chat functionality. Um and I don't know what this this is all about. What's really interesting here to me is a lot of these are new terms to me and the way I see it I see myself as a junior developer you know still learning like uh and I see these as like actually very interesting um opportunities to learn like I'm being exposed to these new terms and I I take them as a way to just basically go on and learn them.

Yeah. Right. Yeah. No, absolutely. Absolutely. And that I mean, you know, no, not not mean to teach us to suck eggs, but that's great what AI is great for, you know, just teaching learning. My my learning has just been expedited exponentially just because of AI. I can really dig into a topic and really query and ask questions about it. So yeah, that's the week's news. Awesome. So to wrap it up, um something that might or might not be AI related, I have no idea. Framer has been creating quite a lot of buzz. Uh a lot of people talking about this. I think there is a lot of you know marketing to learn from them that they are doing that share your screen.

Amazing marketing. Yeah, I'm sharing my screen. Okay, cool. Uh join us May 21st for a day of shipping. Uh that's what framer is saying. I have no idea what it is about. people are there are people who have seen you know what framer is going to be ann to announce and people are like going crazy about it so I just here said now framer is going after Figma by adding more design functionalities you know basically going the other way around but maybe vibe coding coming to framer maybe an actual AI builder maybe not AI at all we don't know I don't know I I asked them to give me a beta preview so I can make a video on it but haven't heard them so someone of your stature Gosh that's rude isn't it they're not doing I'm going to take it back they're not doing a great marketing yeah that could be quite interesting I feel like there's someone else who's launching something soon um maybe It's may I don't know one of these no code tools but there's a lot of announcements.

Everybody's shipping everyone and I'm excited to see what this I've got to go meet a friend now for a drink but I'm I'm excited to read this uh AI chibiT open AI announcement uh that happened whilst we're lolly gagging around talking about the news. So uh let's wrap it up here and thanks everyone for joining in. Yeah. Thank you everybody and see you in the next uh episode. Hopefully next week. Next week. Yeah. Five at five. Yeah. Cool, man. All right. Let's let's end it there. Thanks for tuning in