#60 Webflow's shakey future, Kimi K3 Weights, Jensen on Opensource, GH Stacks

local like like an owner of account needs to be present for every single live stream. Like I cannot believe they haven't ran into this issue in the past. Like surely there are much bigger productions than us who the owner is part of some marketing. We should we should tell them. I have I have and uh I think they I think they responded just saying thanks and the support team was telling me well they're not responsible for Yeah, that's okay. We'll build our own soon. Build our own. Yeah, we're live. Yeah, we're live. 43 seconds already. Time flies. We're live. I didn't know we are live. Since when? Hi. Since um last Wednesday, actually. Yeah. Cool.

Everything you've been doing is has been out there in the world. And so that's nice. That's nice. I should be I should be uh um Are you uh potentially too quiet? Well, that's what I'm doing right now and because we need to check uh on YouTube if the sound is good. Hello, beautiful people on Instagram. Whose uh YouTuber is this like this phrase like check one two. I'm looking at the the what do you call it? The input levels as I go. Wow, that looks like it's really quiet. That should be fine. That should be fine. Let me look on actual Oh no, my USB. My USB is tangled. Hang on. I'm going to go dead for a second.

We should be back now. Oh my god. You're test. They're so loud. There is a noise. There is like a constant. Okay. Test one, two, back, right? And now I'm back. Test test one, two. Test test one, two. One, two. One, two. One, two. Test test test one, two. Right. Let me have a look at what's going down. What would you call uh Timothy Ricks? He's the biggest influencer in the community. I don't want to use the word uh well known well-known creator in the Web Flow community. Yeah. Thank you. Right. It sounds like the audio is pretty good, too. Cool. We're good. We're gravy. If anyone notices anything different, then let's go. What are you writing anyway?

Oh, you just writing intros. Yeah. Yeah. Uh, you could hear it. I can hear everything, mate. I can hear everything. Yeah. Yeah. Yeah. You've been you've been spying on me because of my last Wednesday. Yeah. Since last Wednesday. All right. Cool. So, let let's jump into it because I have a semi-h hard stop and Yeah, me too. And we have a we are getting good at this. We we know we have like a bunch of stuff to talk about and we'll do it anyway. Um so not a major week this week. Not a major week. No, this week we have some big news uh on the web flow side. Web flow no code the whole community side.

Big news. We have a new version of the MCP technology. That's interesting. There are bunch of things that I want you to explain to me and yeah so basically I'm using this show to learn from a real developer to a fake developer a designer trying to be a developer because he's not a good designer anyway. Uh and GitHub stackpull request is interesting. I haven't read into it but you added this and uh I can't wait to know more about it. Anyway, this and a few more news uh in today's episode. This is Command AI. I'm Kabarza. And I'm Sam and this is a week in AI design and death. Wait, is that That's not your mic.

That's your light. Using your light as a mic. I was holding You heard me though, right? Yes, I I did. Any Yeah. All right. We're back. We're back, baby. So we we can like before going to the K Kimmy K1 I want to to mention something interesting like in the shift that we see in the work that we do design and development. I got a new client today. I just told you like they they gave me the green light today like this morning. What's really interesting that uh we offered them a full package of design and development and they declined it. Uh they don't they didn't want to do like the whole design of the website and the whole development.

What they want though is a tasteful team, a a team with taste to come in, design the homepage for them, not the rest. Just the homepage just like the it's not even the bare minimum, the key features that that you know make what makes or breaks the brand. Design that and not even develop it. You know why? because they want to take our design and give it to AI and try to see if they can design the rest with it and if they can develop it and they are very transparent about this this and I am transparent about it and I I'm not like offended or anything. I'm like if that's where the industry is going I mean that's what I'm doing like I'm taking a design and I'm for a new page I'm not designing it manually.

A lot of it I'm trying to extrapulate the design by using you know what already is there with AI anyway. So I find that interesting that we got a significantly smaller portion of the project but it's going to take less. It's just they want what makes it nice and the rest they want to do it with with AI. So, let me understand. They they want you to design the homepage, but they're going to derive the rest of the pages from that homepage. Yes. Yes. Because like once you have the the looks of a card, you know, look at it like an like atomic build or like if you have a molecule, if you have the molecule, you know, like the atomic parts like you know the corner radius, you know the colors already, then you can ask AI to do that for a different car.

So it's not literally duplicating the design. It's kind of like adjusting it depending on the new type of content. The question is, do you need a designer to do this or can you give it to an AI? If the designer with taste already has done the groundwork of the creating brand, maybe this is where we are. I mean, uh, the only that that is fine from a visual standpoint. My only thing is is that they don't you you still need someone who knows what they're talking about to to spot where it's going wrong and what it's maybe doing. Yeah. Where it's going wrong, basically. So, you're not just you maybe getting the aesthetic right, but are you getting the conversion right?

Is the is the web flow is the web flow. Is the website built in a way where it makes loads of sense or is there a bunch of boooos? So, there's there's more to it. very presumptuous they have a no in this case they do have a good UX designer somebody who has made the wireframes already so the wireframes exist the UI doesn't what we do we dress the wireframes not all of them actually we dress one wireframe and they will see if they can dress the rest of the wireframes with the same UI we will find out uh I'm very like open about it Um, and I even offered them QA. I said like as a part of we can extend the package where you do the work with AI and I will point out what might be wrong with it and I get paid for that.

We'll see. We'll see. But it's interesting. Very interesting. Cool. Chem K3. We diving into that. Yeah. Cool. So, the world's best open-source model going toe-to-toe with Frontier models like Opus, like uh GBT 5.6, uh finally released its weights. Now, it released the model first. Now, it's released its weights. Finally making it something you can run at home kind of. We'll get into that, but point is if you have a supercomput, if you have a supercomput, we covered this. You know, this is where I'm show where we can see us going toe-to-toe. And they said they would release the model, the actual weights that create the model. This has taken a long time, the screen share.

I'm sorry, guys. It just for some reason just takes Yeah. Um, it did have it. Maybe they've updated the article, but it it came out. They finally dropped the the weights last week. We've already covered Kimmy K3. We don't need to talk about anymore. But what does this mean? What does the open releasing the open weights mean? Well, first of all, it means if we go to um open router here, it means that we get all of these providers able to provide the model for different prices. Now if we look where it is it here fireworks fast is slightly more expensive about like you know 2/3 more expensive but the speed and latency is a lot better so the tokens per second it generates a lot faster and the actual latency in between calls is a lot faster so they can afford to charge a lot more there's nothing below the uh RRP let's call it um but this is what it means that people can now run this hardware.

One thing I did want to call out is you're paying $3 and uh $3 in $15 out for open uh digital ocean, but the latency is pretty poor and the tokens per second is pretty poor. Why would anyone mention that digital digital ocean? I don't know. I I wanted to ask like nine tokens per second. That's that's really low. That's awful. Yeah. So, but anyway, the the point is is that now we uh now we can run it here. What that also means is if we go on to well, let's go to Onsloth here. Onsloth is a uh I don't actually know okay train and run models locally. I I see them all the time on hugging face, but they managed to run quantise the model down to one single bit, which means they shrunk it from 1.56 terabytes to 594 GB, which means it can run on a Mac Studio with 512 gig of RAM plus 125 gig of RAM device.

So, still very much out of the wheelhouse of many many many people. However, this is pretty uh let's say um obtainable if you are if you're willing to to commit to it. What is the token per second for that? We don't know. I don't know the token per second. Um maybe we I'll go on to the page in a second, but actually we'll go on to it now. So, let's have a little look at this page here. And just just for comparison, uh GPT 5.6 Soul uh achieves up to 750 tokens per second. Um that's not what we get. We get probably closer to just 50 just just to have a comparison. Yeah. Well, to be fair, it will obviously depend on the hardware you're running it on because you could have a your 128 gig device could be an M3 Max or something like that or it could be an M5 Max.

Do you know? So, the tokens per second is is completely dependent on the the hardware you're running it on. However, the interesting points to note here is while being 62% smaller, it reaches 90% of the accuracy. So uh hang on. So this sorry reaches 72% top while act. Okay. So this is very very confusing. Dynamic one uh is 78% smaller while be uh sorry 70% 78% my god this is awful. 78% more accurate while being 62% smaller. Dynamic 2 uh 90% uh accurate while being 45% smaller. So you're not, you know, quantization is this goal of running um or shrinking a model while maintaining as much of the intelligence as possible. There are many many way many ways to do that.

Uh but yeah, the this is what they're running. So, let's have a little look if they have a tokens per second thing on using their recommended Max Studio plus um uh let's actually just search TPS. No, I don't think they got it. They haven't got it. Um but either way, pretty cool. I actually originally thought this was 12 like you could run it on 128 gig thing and I whacked on my camera like set up my setup and I was like hang on this is like 500 gig I can't run this on a 20 128 gig MacBook but yes and final thing to share is that when you go on hugging face these are all you could you've got the original weights here you've got gguf version which is um a single package version which can run on max to be fair but this is going to massive.

It's one 2.8 terabytes. Uh but you've got um for DJ DG Spark or DS Spark, whatever it's called. Uh here we go. That's what I was looking for. So different quantizations, different trainings, people have tweaked it. People have done all sorts of nonsense through it. This is what open Weights is all about. So yeah, still you can run you can run it at home. Um, but you still need some hefty hefty gear. The only other thing I was going to say is, uh, I'm going to quote myself here. So, I was thinking about this the other day that like Kimmy K3 is is amazing. Like the the fact is, uh, if I show you again, just remind you, like the fact that it's going toeto toe, as I say, with all of these models like it's next to Fable and GBT, uh, 6, 5.6, six, all that sort of stuff.

The fact that it's doing that, it is incredible. And these are like freeto use models. There's no clear uh path to monetization on something like Kimmy K3. So, but on so on the one hand, I'm like, you know, this this just shows that um basically the Chinese government are willing to fund this because they are the Chinese government are are subsidizing the training and whatever of these open- source models that we're getting for sure, probably some um other investors and stuff like that that it takes a lot of money to train a model. So, there's definitely a commitment there. However, like this is a massive model and opensource is all about not necessarily all about but it's all about flexibility and freedom and being able to run it locally and whatever.

And I posit the idea in this little message here is if GLM or Quen or Gemma, one of the other kind of running models that you can run locally on your machine, if they were given infinite budget on like, yeah, make this as big as you like. I wonder if they were a would be able to uh achieve the same sort of performance because this is this is doing really well but it's also massive. It's like what would be a good analogy? It's something to do with money probably but of course like if you plow more money into something it's going to get better. You're going to get more out of it. So like I don't know.

I I really like Kim K3 and I end it by saying I don't dislike Kim K3 and I like open source, but I wonder if the fact that it's so high up is distracting from the fact that actually we don't we lose a lot of the benefits of open source purely by it being so faking big. I mean, they're not necessarily charging less for it or, you know, we don't get tremendous um you can get Lama now I think as well. But I don't know, it's an interesting one. All that to be said, Kimmy K3 is now open sourced. Uh, you can go download it, you can do whatever you want with it, you can train it, and if you have good enough hardware, you could potentially run the one bit quantization version on your Mac Studio plus some other device.

So, pretty good. Yeah. Nice. Um, we we have got some more open-source news. So, should we stick on the open source thing or should we mix it up? No, like let's let's go with open source. Okay, that's good. Let me do it. So, Jensen, do you know you know Jensen, the CEO of Nvidia? He's joined Twitter with a leather jacket. With a leather jack. Yeah, he's the cool guy, right? um he's joined Twitter and he made a bombshell of an announcement or at least he's said something that's causing a few ruffles of a few feathers which is basically an open letter talking about the importance of open weights in open source and open weights just in general and uh the world needs both frontier closed models and Frontier open models.

And this has led on to and unsloth here is making an appearance. This has led to a bunch of companies including Meta, Microsoft, obviously Nvidia, Google, all of these major major players. I think even OpenAI is on here somewhere. Say when you see it. Olama, Planter. I see it. Next, Open Claw. Yeah. Oh, yeah. There we go. Yeah. See massive massive names on here. uh all supporting open source and of course as you know open AAI do have an open weights model or open source model sorry not open weights um but they are primarily closed source there is definitely an industry for both who's missing do you think on this little list here do you know who's who's missing on this little list here where where is our boy Dario Dario where are you anthropic Where the hell are they?

They are I mean they're not here, right? They are not here. Really? Even Palanteer is there. What the hell? It's It's definitely um it's definitely raised a few eyebrows the fact that uh Anthropic aren't on there. Uh and it's sort of it's sort of expected but not expected at the same time. Do you know what I mean? Like it's like you thought they would at least have a go. Um but the other day they released our position on open weights models. Now I um haven't read this uh I haven't read this uh thing but I'm getting my AI to quickly summarize it for me on the right hand side here. Um core position anthropic states explicitly that it has never advocated for a ban on open weights models and does not see uh protectionist bans as useful.

So it's not necessarily going to ban them. Open weights model without dangerous capabilities are framed as public good offering free widely accessible value to business developers and researchers aside from influence and compute. Their main concerns are national security and that authoritarian regimes particularly the CCP. I don't know who the CCP is. Do you know who CCP is? Yeah, that's uh China. China. Oh yes, that's the China's Communist Party. Yeah. Fine, there we go. Could develop more powerful AI models than the US and use them for permanent military superiority and deep domestic repression. A secondary concern is that powerful models, open or closed, could be misused for cyber attacks, biological attacks, and exhibit serious alignment failures with open weights posing higher misuse risk due to lack of enforcable guard rails or monitoring.

So basically they're saying that I mean AI should be in the hands of those who know how to wield it in some sort of ways. And in some ways I don't know in some ways I sort of agree but in other ways like wow look at look at what we're talking about the open weights I mean even even uh China's or even even open weights open sourc's best contribution to date is not just not as good as um uh fable or GBT 5.6 or anything like that. they they're going to they'd have to they'd be really hard pushed to um get a model that is capable of like warfare and whatever, but it's not impossible to do so.

Um, and it it it reminds me of what you covered a couple of weeks ago when you said about the um GBT 6 breaking out of the containment and hacking hugging face and then having to use open weights models to bypass the um restrictions that Fable or GBT 5.6 whatever they had access to was placed on them. So they had to do it. So these these these models are capable. These open weights models are capable. So yeah, that's kind of it. It's kind of funny. Uh yeah, they are scared and I I like as you've been reading through like these and explaining Kim K3 and also this open source thing. I've been thinking like for the things that I'm doing, do I need the most capable model all the time?

And I have been using, you know, I've been using Fable to start the server, but I'm thinking maybe not. You Mr. Max, Mr. Max everything all the time. 100%. Fable X High start the server. You've changed, man. What's happened to you? You've changed. Well, you you've been you've been influencing me. I need to go. You've been influencing me. I I didn't even used to use the terminal. Now I live inside the terminal. Um I used to max everything and I'm seeing like no points like for like things that I'm doing when I create a skill after like trying it with Fable once. Maybe for the second time if I do something similar or the same thing which happens a lot.

Maybe you don't need the the craziest like strongest model that is going to cost you a lot of money and time and a cheaper model could just do the same exact job, maybe even better uh and do it faster. Which one would you prefer? Wait 2 hours for Fable to cook something really good or maybe 5 minutes for Soul to make pretty much the same thing and then you get into, you know, the feedback round. So this is like more the exact workflows that I that I'm talking about like yeah like I I would like to so what I'm doing uh is using fable to plan and using other models not on high like I've been using opus and sonnet on medium and low because if I have the plan sometimes I just need a model without a brain to go in and execute.

I don't need I don't need thinking too much. I don't need to like the feedback the feedback that you get quicker is often more valuable than waiting for the perfect solution. It's agile, right? It's it's that mentality. Exactly. Exactly. Because kid you not like Fable can cook for two hours and then you like you're sitting there like waiting for two hours. like there is no point at that for so many things that you do and yeah as you said like I yeah changing I've been using soul a lot uh the 20 bucks subscription does feel like the 200 from anthropic maybe it's not the same maybe it's like the 20 bucks uh subscription from OpenAI is closer to the 100 that you get from Anthropic which is crazy.

I'm considering buying uh GPT for my team, like GPT team subscriptions because that's like economics and we don't we don't need the craziest like fable all the time every day. That's right. Um I mean and we didn't we didn't know whether we would cover Opus 5, but I mean this would be a nice place to kind of discuss Opus 5 and because it falls into play with all of what we're talking about. So I I don't mind Opus 5. is it's it it does the job. It's pretty good. Um I think it comes down to just understanding the task at hand a little bit and just knowing is this something and and knowing a little bit about architecture of code like is this something that touches many many files that it's it's a complicated thing it's touching uh the database it's also it's a backend it's a front end it's uh access whatever all of these things is it a big task then you're likely going to want to reach for fable or opus and go on high mode and potentially even plan.

Like there's a debate at the moment whether plan mode is necessary. I do think it's necessary. I don't want it just to pull the trigger without me just reviewing what it's doing and and you know I just want Yeah. Interesting. Why would you use the plan mode and not just say create an MD file? Well, that's all that's all plan. Oftent times I don't create MD files. Ive I've I've kind of moved past that unless there's a there's a few scenarios where I've asked it to create an MD file, but most of the time I don't ask it to create an MD file. What I do do is say create this but use sub agents to then then that main chat where the plan was done actually just you know teammate sub aents on it.

Um yeah. So yeah, with that understanding the task at hand, um honestly with smaller tasks and realistically, you know, with let's say Zoku, the website we've been working on or let's say um there's an app that I'm building at the moment which is a really really basic like you know it's it's complex in that the work has already been done. So it took it took you know Opus 4.8 when I built it you know on high to actually build the application that it now is. We're now making a web version of it. A lot of the architecture a lot of the decisions have already been made. Now I'm adding additional features to it. Um Opus 5 on high was just taking so so long and I dropped it down to Opus 5 on medium which was a lot faster.

And fun fact, Opus High uh Opus 5 on medium is the same as GPT 5.6 high. Uh and it's cheaper. And it's cheaper. Wait, Opus is cheaper. Opus medium is cheaper than GPT 5.6 Soul on high. Interesting. Interesting. Maybe I can show you it so we can see it on deep. It's on deep suite. But this is interesting like but it's kind of like yeah we are getting to this point where we are using smaller models because we care about speed. I didn't before but uh I you okay this is embarrassing but I used Fable 5 to install in static I didn't know I didn't know I didn't even like read you know how like how I'm installing like I just told it to install it today I installed it again for another project without without any model I just open the terminal in the directory I want I just copied like pasted like get clone, you know, I know what's happening.

Uh, and with Fable, I think if I'm not mistaken, it took like 36 minutes or something like it's crazy. Just like it takes literally 5 seconds. It should not like you don't need agents for everything once. But it is interesting because we are I'm saying interesting so often. I'm getting as I'm getting more into this work and like doing the development stuff, I feel like I'm getting more comfortable and there are points where I ask, wait, what did you do actually? How how is this working? And maybe I do it myself next time. Yeah. Instead of waiting 30 minutes for a git to get cloned. I think definitely remaining inquisitive and just not mindlessly just chewing through your a your tokens like just getting stuff done.

Um so this is what I was talking about here. So you've got medium. This is o this is opus on medium. So it's marginally it's marginally cheaper um and marginally less intelligent. So you're basically getting performance uh the uh 5.6 high performance on medium. So, that's just something to bear in mind. Um, but yeah, given that, all that to be said, having it on Medium for some basic front-end task, understanding what it is that I'm doing. I know I'm not touching a whole bunch of files. This is not a complicated project. You probably do not need uh the crazy models. I don't think I can drop down to Sonnet. I I just think that Sonnet is just a bit of a weird spot.

like I'd rather go and in fact look at the you know if even if I was to go to um Opus 5 on low it's still I mean which which one's more int which which one's intelligent it's still much more intelligent much much much more intelligent for cheaper than if I'm reading that right do you know what I mean than than sonnet even on X high if I'm reading that correctly mean is it like price per token or per task because I think I'm not too sure to be honest. Let's have a look. Uh that's the thing. I mean this is the task cheaper tokens, but you might need more tokens to get it done.

Yeah. Yeah. But either way, like I'm not looking for I am and I'm not I am like I I don't know how to explain this. Sorry, I should have been sharing my screen there. Um this is this is Opus high on this is Opus 5 on low and it's got 58% score at $166. Whereas uh high doesn't reach that. Sonic high doesn't reach that and it costs $12. So like that's what that's what I was looking at here. Um I need to bear in mind that my job is to create apps and websites, right? I make like I make good money off of doing what I've do what I do because I've been doing it for so long.

I think there is a lot of noise. I I'm I'm probably going to say something uh pretty hot hot take right now. I think there's a lot of people making a lot of noise online who either don't are just hobbyists. They're looking for the cheapest model. They're students. I It's hard to pick apart who's complaining about the price. And my gut feel is that a lot of these people care about price who aren't earning a living. Whereas I'm learning a living. like I will pay whatever it needs to get the job done because the the opposite of this is me hiring a developer for you know a ridiculous number of a ridiculous amount more to to have a developer do this whereas it pays for itself every job pays for itself you know I could pay for a year's worth of um opus uh claude anthropic in one job right so that to me is just so so so worth it.

So, all that to be said, I I don't know whether price is really an issue for me. However, I do know that correlates with how much usage I could get out of a month. And to be honest, I would like as much usage as I can get. Um, so I still balance out using Opus on medium versus Opus on high. I don't think X high or max is worth it. I've done it a couple of times uh to go to X high, but uh yeah, that's my that's my hot take. The internet's full of a lot of hobbyists and students who are a bit tighter with their money, and they need to be. I'm not saying they're wrong.

They need to be tighter with their money. Yeah. Uh I see that. I I also see that big companies maybe they've been spending maybe too. We we've talked about this in the past. Mhm. Um noticing that the cost is too much for the perform the the output they are getting. But I don't think that's a necessarily only the issue with the cost of AI. Maybe we just don't have the processes and like the understanding of how uh a corporate a company supposed to use AI to gain efficiency or like more output because like if you give people just use AI they will use AI to write big emails and then on the other end they will use AI to read those emails.

Yeah. So it's like I mean sorry but it's it's no surprise that enterprise just are oil tankers who leak money you know they just don't they don't know how to be efficient right and that's why it's costing them a lot of money they also don't get access to um claude max subscriptions they have to pay API pricing I think they can use codec subscriptions but point is it's it's not for me it's not uh it's not relevant it's not relevant a talking point because I'm a solo developer and this is what works for me. So, I just want to make my position known uh as a solo developer on a $100 a month plan, not really being too mindful of price just because it makes it money back.

So, yeah, it's um I How did we get on to Opus 5? I think was it Jensen? Was that our boy Jensen who got us eventually on to open sour uh uh thing? Yeah. Yeah. Well, there you go. That's that's the that's the beauty of a podcast, huh? Or whatever this thing is. Whatever this thing is. Um I'm trying to think as we traverse through all of those different subjects if there was anything that popped up into my head on the on the way. But uh I I guess I'm thinking more as well just about yeah just swapping swapping um out different models for different tasks mostly because it's faster. I did find Opus 5 to think a lot like it would take a long time for like marginally worse output than Fable.

So I should have just used Fable for it. There are still probably some use cases for fable but um yeah I don't think design work needs opus you know visual front end design work I don't think that needs opus what do you think for designed fable fable is just so good at design I sorry I don't mean um yes what am I trying to say here I guess I'm talking about front end development like actual I'm not I mean I know you once you have the plan Even Sonnet is fine. Exactly. And that's why I feel like um the the the tasks for front- end work is Sonnet is is mostly fine or or Opus on like low.

It's not it's not complicated. I think understanding where complexity really lives uh in before you start the task is you know if you understand how apps are put together and how they architect I think I think you'd start to realize where complex task what complex tasks really mean you know are there cyber are there cyber security issues are there stability performance issues and and all that how many files are they touching like what is the complexity of this task if it's just add this slider to this page page, you're probably going to be able to get it done with thing because it's only touching a couple of files, you know. Yeah, that that's actually a really good point because this is where I understand that I don't understand enough.

Mhm. Does that make sense? This is like where like if we do something with Webflow and I'm doing something with the MCP. I did like last week uh I updated a client website with the MCP uh and I asked it to write exactly the steps that it did in a table because I can just like skim through it and I can I understand like how the CSS in web flow works, how like what it should do, what it should not do. But when it comes to a complex web app, I'm like in in the mercy of the AI knowing what's the right thing. And sometimes it, you know, asks you question like between these two technologies and architectures, which one should I go with?

And I'm like, mate, you're like you, you are the driver. I have no idea. Here is here's something I thought of this morning because I'm I'm telling Opus to like chill the out with trying to overly explain things to me. And sometimes, even as a smart, young, handsome, sexy boy like me, sometimes it's still a bit too above my head in its explanations. So, I I've I gave it well, not that I've given it, but I prompted it to say, can you explain this is a bit dumb dumber for me? And it was code. It was architectural, right? I think people should put in their root like agents or Claude's MD file your level of understanding.

Explain to me like I'm a junior developer. Explain to me like I'm a mid-level developer or you know be in control of how it narrates its work and what it does and just set that makes sense. Yeah. Yeah. I thought that I thought that this morning. Um because yeah, sometimes it does and it tries to because I've been doing a lot of marketing stuff recently and it tries it just tries to sound a lot smarter than it needs to be and and it tries to be cooler than it needs to be, you know. Um I was talking to you before the show about the conversation I was having about my business and it was something like it it's not the um I've forgotten what it exactly it said, but it ended with it's the shape of your business.

And to me it was just like that's a really bad explanation for it, but ultimately it just phrased it in a way it's like just say it what it is. Like don't try and package it up in something that sounds interesting or intelligent or like cool. Just tell me what that means, you know? So anyway, um yeah, I think we're start we have to tame the models um a bit more than what we used to. I can't speak for soul, but definitely Opus 5. Just while we are on pricing and all the rest of it and we and just cuz we haven't really got a story about it. Nice to know that they've made Luna 80% cheaper and Terra 20% cheaper, which is pretty cool.

I think this is going to really help people who aren't coding and doing complex tasks and things like that who are using chat GBT to like maybe more geared towards work maybe I think GB chat GBT work um I think people are going to really the sort of lower to mid-level kind of aentic I don't know work people blue collar I don't know what you call it going to really appreciate that and then there's a fast option for GBT 5.6 six soul in the API. So if you use it through the API, you've got a fast option uh which probably consumes obviously more probably the same price but because it consumes it a lot more fast.

Maybe it's double. I don't know what the prices are but yeah nice little bit of news there. What's next? Should we should we go on to web flow or we can can go to web flow or or anthropic just because we've been talking about anthropic. I like I like smooth I like smooth segways. Yeah, let's do anthropics. Uh but I need to pull up the uh so maybe you explain this one because even though I added this, I don't understand it. I haven't read it. So maybe I'll need to um tell me what you know about it and I'll read up on it just in the background here. Uh, do you know this alphas signal.ai site?

No, but it looks nice. Okay, so let's actually I I need to The graphics look nice. Yeah, but I think uh in order to read it, I need to sign up. Oh, I can read it. I'm reading it now. Oh, no. Actually, yeah, it fades away. It fades away. I'm going to sign up because the design is nice. I'm I'm going to trust them. I do think it's funny. I think it's AI code. I think it's AI designed, by the way. Yeah. Yeah, it is. It does look like it somebody taste. Yeah. If you hover over the left side of the page as well, I can maybe even just show you whilst you're getting that ready.

Uh if we go over here, it's like a nice little I don't even know what it is. Oh, it's the Oh, it's the um chapters. Yeah. How much is it? It's free. Oh, okay. There you go. You can sign up then. But yeah, these graphics are quite nice, too. Okay, it's AI coded. It doesn't let me add my first name or last name, but it doesn't let me move um forward with the form because it needs Yeah, whatever. It's It's a bug. It doesn't work. Uh, I hate that that it's like I can't even sign in or sign up. Uh, let's let's see if it was successful. Yeah, it was successful, but I couldn't make my account.

Okay. So, uh, I will show it in a second. I think I understand it. I'm just reading the TLDDR at the top. Yeah, you should like I think I understand pretty much none of none of it. Just uh let me uh give me a second. Am I back? You are you back? What do you mean you never went? Yeah, I was gone for Yeah, I was like disconnected. uh my headphone. Oh man, it's so hot. It's hard to concentrate. Um Okay. All right. So, you have probably heard of MCP and there is a new version of it. Uh um Okay, maybe I'll go again. I should have called that one, shouldn't I? Yeah, it's like it's on the dot.

Okay, so there is a new version of MCPS and this one is MCP as stateless. What the hell that means? Um, I don't know. Sam will explain or we'll try I'm trying to shift this explanation on him. But uh apparently from what I understand the way the MCP works when you when you tell your agent to do something to update somewhere it like there is a list of things to do and it does them like one by one but uh it has to like wait for a tool or like a call to end and then it can do the next one. It can't do a bunch of things at the same time. I might have been not completely right about this uh or like technically correct but that's how I understand and with the new one it's stateless in the sense that it doesn't have to wait it like communicates faster.

Can you can you like explain what that means? And somebody by the way mentioned we are full circle back from you know developing MCP as a new thing to now just being to have REST API. You know what rest API is, right? Yes. Yeah. Yeah. Yeah. Uh do you want to uh zoom in just a little bit and I can kind of explain? Yes. So the way that um uh MCP's kind of worked before, remind me to talk about the what did you just ask me? REST API. No, let's go with this. And then remind me about REST API, right? Um so let's go step by step. Yeah. So before as you can see by these uh images here when you have an MCP it's running on a on a server all the time you need a node server or a Python server to run the and accept requests from the the whatever it is accessing the MCP the agent or whatever it is right and what this means is you got to think of it like a motorway a single lane motorway and you the one cars.

If you want a return journey, the one car's got to go to its destination and then wait till that's roads clear before it then drives back, right? And that that I mean it's good because it can maintain a session. You can maintain knowledge between um the the state I guess you you call it. There there are benefits to that. But concurrency which is multiple things firing off at the same time becomes a lot it's impossible, right? It's a single lane as you can see from this image on the left. If you go to the right, what they're saying now is that MCP is moving to a stateless mechanism where one you don't need a node server constantly running to uh serve the MCP.

It can it can start from a cold start which is what you call it which means that you can deploy the MCP and it can scale depending on the number of requests it the load balancer that thing in the middle receives. So if one is busy if instance A is busy doing one thing then load balancer can say hey instance A is busy I'm going to route you through to instance B and if A A and B are busy then it will route to C. If A, B and C are busy then it can scale up to D if it needs to E FG or whatever. So it just enables it can be a lot faster.

I think there will be some downsides to this but in order to maintain session data to you know they say it's stateless. Sometimes you probably going to want a few things still maintained in state. It probably the shape of the MCP object that gets returned is probably got some additional information on it that might you know if it's say I don't know what I've never built like an MCP to this level. If say you want um a a tool called or or or a prompt even to get sent through and not get lost, then it might return what the original prompt is that you get sent through to the to the next instance because it's completely stateless.

Doesn't have a clue what happened before or after it or anything like that. Interesting. Okay. So that's it. It basically comes down to speed and concurrency if I understand. So is it up to if it so based on that it's up to the agent to decide or like to understand the state like has like because like if you want to do so who decides that like if you want to do something often you want to do something first and then something else later. Um well it I mean it's more so that it can do several things at once. So like it can it can fire off a bunch of tool calls to the MCP and get them done simultaneously and different calls.

Um I I suppose I know what you mean when you say who's in charge of the state. And yes, I guess like the agent needs to receive the information and then send the relevant information on to the next uh instance if it if it so needs it. it, you know, your MC your your agent might not need the concurrency and speed, but at the end of the day, if if it's available to it and you need to build something around something that does require the concurrency, then it has that option to kind of do that. So, yeah, I'm not sure about like I've never been so concerned with the inner workings of an MCP. I've built a few, but my agent has built it.

So, it's it sort of makes sense from that perspective. But I would need to read this article to really understand where the pain points are and where this falls short and where it really helps because in day-to-day it might not help dayto-day. You know what I mean? It might be advanced MCP setups that this really comes in handy. I don't know. I'll need to read the article. Okay, cool. Well, uh, from what I see, it's also a proposal or like is it it's not deploy it's up to the No, it's it's final. It's final the specific. It's final. But like for example, for the web flow MCP, it would be up to web flow to uh update the MCP to be like that to to use the stateless design, right?

Um, I believe so. I'm not sure about what the what this implements. I think it might just be that uh it might just be automatic. I don't know. Um but I guess you'd need to deploy this MCP on something. So yeah, I guess there would need to be some infrastructure change. Yeah. What was your question on REST? So REST API like what what is REST API and how is it why is it compared to REST API? So um why is it compared to well MCP is just I don't know what it's referencing in this article. I can I can only answer your thing on it on face value but um an MCP it starts with an API which is why when we've talked about web flow in the past it's like well whatever the web flow a API can do the MCP can do.

So you start with your API and there are different ways to create APIs. One is restful um which is a certain kind of protocol. I think another one is SOAP which I've never touched. I've never done a SOAP API. I think technically GraphQL is a REST API. I'm not 100% sure on that one. But the point is there are different ways to interact with backend data or you how you create the the API. Mhm. The okay MCP is a layer kind of in front of the the API, the REST API that just reveals what the API can do and the shape of the request that that API needs in order to interact with the endpoint.

So it's again just a layer in front that the agents can talk to. So, it's nothing really more than that in terms of how a MCPs fit into all this, but just know that a REST API is a is a way to interact with uh a backend and and data databases basically. So, there's a bunch of endpoints. You got you got often create, put, delete, um the Oh, there we go. Yeah, it is SOAP. There we go. Create, put, delete, and post. You know, you know those classic ones. Those are the And then and then you could have you could have slash um command AI um uh post and you can get a post you can you can post a post which is creating a post you can update a post or uh by passing an ID.

So it's all the same endpoint but it's the method that changes that determines which method gets called on the rest API. Yeah, GraphQL is an API. There we go. I wasn't I wasn't far off. Websocket, Web Hook. Okay, nice. Yeah, I've heard of these, but I didn't know that these are API. Exactly. Neither do I. Not neither do I. I didn't even know I didn't I didn't know WebSocket or Web Hook were technically classed as APIs. So, there we go. But interesting. Uh this is an article on Medium and it's called what's API API types, most popular API services, REST versus SOAP. what's the difference? And you know, again before the call, this is why I asked you when we talk about the Web Flow API, and this might be a good segue into the Web Flow article.

Um, Web Flow have released their Web Flow 2.0 um MCP, and I'm so so I'm curious what features are part of this. Obviously, maybe they've um updated it to the restful and that's where they're getting their performance out of or yeah state stat stateless sorry that's what I meant the stateless maybe maybe web flow MCP has literally just updated to this new protocol uh to get that performance and speed however it's just a wrapper around the API so I'm really interested in what new endpoints have been released with the web flow API that make web flow MCP 2.0 O uh so great. So I haven't looked into it yet. Yeah. Um it will be in the article like not exactly the answer to what you are asking but somehow Oh nice.

There we go. Uh I noticed that we have some comments and like I hate that Instagram you know doesn't show me the comments properly. It it's a mess with Instagram. It is an absolute mess because I see everybody who joins and the comments get lost. Uh anyway, so it's probably too late for the comments. Uh before we go, we had also a comment here. This is like at the start of the show. It's kind of funny. Yeah. Here's a question to our audience. Uh, I was just thinking we have currently got, shout out to the zero viewers we currently have on YouTube right now, right? We don't get a lot of viewers on YouTube. However, it's more about the the actual clips that go out on YouTube.

I'm wondering because we only have a limited number. We're not we're not making loads of money yet, so we can't go up to the next level up of Streamyard. I'm wondering if we sack off YouTube and just stream to my Twitter, your Twitter, and your Instagram. Yeah, cuz I think those got those get the most engagement. No one on YouTube, no one on YouTube is here to like, you know, shout at me for even suggesting such a thing. But proof is in the pudding. We don't actually get that many. Oh, we got one person. Thank you for joining us. I wonder how you'll um how long you'll stick around for, but we've got one viewer on YouTube.

We're considering not streaming to YouTube, but shout out to you. Give us a like and uh a subscribe if you uh if you enjoy the content. Anyway, what are the comments on Instagram? Uh they were asking like about how to use their tokens more efficiently and jobs that are more what's the most important job nowadays for uh p like for someone between 25 and 30 and I would say like from the the things that I'm reading and watching a lot of companies are rehiring they they fired a lot of designers and developers like mostly developers and they are rehiring because they thought the AI would do the job alone but now they notice it's you know you need this the same people but maybe just with different tools maybe they are more efficient with these tools maybe not depending on the job but like my bet is on anything like engineering or um like STEM we know like like these AI models are really far away from the AGI that we were promised to.

And even some altman admitted like it was last week or so that he was wrong about the job um lost like that millions of people would lose their jobs. That that has not been the case. And the companies who have laid off people, they are hiring back people because they they need the people. So yeah, I think you still need someone who knows what they they Yes. knows the the job, right? So, yeah, I don't know, you know, I don't know. I don't know about engineering. It I do have my doubts about it because right now, you know, those lower entry-level jobs are going to get sucked up by people who are really good with AI.

I think that they're hiring people back, but they're not hiring their whole team back. They're just hiring one extra person. They thought they thought the design team could take care of the engineering, but it's like, actually, no, they can't. So let's hire a developer back to take care of the AI to manage the AI to do the development. I think it's more like that. But no, like there are like big like they there were companies that fired a lot. I think Ford was one of them. Like they fired pretty much like all of their like engineering staff in a department apparently and now they are like looking into hiring pretty much everyone back or like a big chunk of it.

It's not like one person can do it all. Like if you can do it with one person or do it with 10 person but have 10x of output, why would you do it with one person? Why wouldn't you half the workload and scale? Yeah. Yeah. Really, it really depends, right? And there will absolutely still be jobs for engineers, but if you're 10xing one, if one person equals 10x, that's nine less developers to uh that you need to hire. doesn't mean to say that you know you want to basically find a balance. So there will absolutely still be jobs, I think. But I don't know. It's it's I think I I I I see you are more leaning towards probably more like jobs will be lost still, but I'm more leaning towards the Jins uh paradox that I I feel like code will not become obsolete.

Actually, code will become absolutely less scary. like very similar to writing like how like more people will know know code. They will like read code. Maybe not read every line they ship but they there will be more people like the focus here is is on more like in the count of people who will end up reading some code uh learning code and touching code. touching code directly, which is also a segue to, you know, the the web flow thing. It's about touching code. It's literally about that. Yeah. I don't know. I mean, it's impossible. It's truly impossible to read the tea leaves at the moment on where things are going in in dev and tech.

Um, my my worry is that contributes to my less jobs thing is obviously the AI bubble factor we're in. Obviously the rising the costs of AI and the subsidization that's happening right now is already quite difficult to manage let alone when they finally have to say when they finally can't rely on stakeholder um uh what do you call it like um investment who knows but job-wise become a plumber AI can't do plumbing not yet uh the the robots But from Elon will no I'm kidding. Um I would go towards like any sort of deep knowledge and STEM and uh whatever you do do what you like do what you like that because this guarantees that you can stay at it and then go deeper and learn more.

Right? If you if you do it but don't like it, you you won't uh do it for long. More people won't read code. Uh matter of fact, they are hiding the ability for you even uh to see to even see code. That's not my point. Yes, I see more like more and more developers are read reading less of the code that they output. But there will be more people who will work with these systems and maybe they start with something like lovable but they will get exposed to code and they will end up reading even a one line of code. There will be more people who will get involved. That's my point. And I don't agree that that they're who who are they?

I don't know. But I don't agree that they're making it harder to look at. No, it's actually easier than ever to like find code. People are looking for the code now. So I don't think that who's who's hiding it? Who's hiding it? No, but I think um the the point here is like I think they are this person you're uh probably mixing new tools that don't necessarily show the code in the first glance with hiding it. They are not hiding it. Yes, there are tools that are trying to abstract away from the code, but there are we will have an article on that. That's that's not going to happen. Like the whole industry is like noticing that they have to build directly on top of code and not abstract away because that's not going to work with AI.

And uh when you build directly on top of code, sooner or later the users will have some interaction with the code. Again, I'm not saying like anybody who's doing any type of development is reading every line of code that the agent is producing, but you are getting more exposed to it and you will slowly understand some part of the code and I think there is fun in it like understanding you know what your code does piece by piece. Uh there will be some exposure to it. Yeah, I'm just signing up to that alpha thing as well because it looks interesting. Yeah. Yeah. Alpha signal. Well, yeah. Um should we get into the web flow article?

Uh, you know, getting to the source. Getting to the source. getting speaking of hiding. Yeah. Uh so I I wrote a I wrote an um intro and I'll try to go with that and yeah so web flow admitted that they are playing catchup. They actually admitted this and we are seeing the community falling apart and even one of the biggest creators in the community Timothy Ricks probably you have uh Timothy Rick you probably have heard of him uh he's falling out of love you would say with web flow he's like using other tools and starting to educate us on other tools and a lot is going on so um we want to talk about that maybe that was not the missed uh intro.

I'm sure we could edit something out of that. Yeah, maybe I could scrap it together. Maybe I'll I'll try one more. Maybe this is a good video. Maybe I want to give it one more try. So, Web Flow admitted that they are playing catchup. The f the when when you do the second time is working. We flow admitted they're playing catchup. Pause. Just pause after that. Web flow admitted that they are playing catchup. The community is falling apart. The biggest players in the community including Timothy Rake, they are ex uh they are working with other um tools and they are educating on us on it. And Web Flow even came with a big announcement well a direction you they call it uh article where they are explaining where Web Flow is heading probably webflow 2.0 3.0 draw if we count like framework does.

So let's get into it. I want to show you a few interesting things. So let's start with this. So this is the article but let's start with this. Have you seen this video? Have like do you remember you you don't this video is so iconic like this person trying to get through the door and the door is broken. The hand is broken like the images are missing. Things are glitching. This is 200 what? This is five years ago. This is I believe like 2019 even uh 2020. This was web flow mocking WordPress. Like the plugins break, you touch something and things fall apart like like exactly this table. Cool. and that there it's a really good video, but it's really sad like this is where you meet Web Flow and where things don't break, things, you know, work as they supposed to and you have a visual builder.

You see everything on the canvas, you change them, you immediately see the change. And it's really sad. I've been thinking about this because this video is true today about web flow. The parts where things break, which is sad. If you use the MCP, you might have really good experience with it. You had good experience with it, but in my experience, things can break very fast. And I noticed that things are not really like fast enough when I do them manually. And if I use like AI, I'm just super fast. But when at the moment I'm using web flow, things fall apart and things are really slow. Not to mention, there is no extendability like with the standards that we have today.

So the video that was true about WordPress today is true about Webflow. And I find this a bit sad because I was having so much fun with this video. I used to show it to my clients. I used to show it to people in the community who were just learning about web flow. I was using this sales pitch. It was my sales pitch. There there are a few other videos like this like how fast it is to do it with web flow where in the old way they call an engineer to change a button and to change a copy they have to call another engineer somewhere else. Now, all of that is true about Web Flow where with AI, you just type it into the AI and it does it for you.

So, time has changed and the community is seeing this. So, I want to start with this post before going into the the main post by Timothy Ricks. Uh, I've been Web Flow's biggest fanboy for the past seven years. To me, this year's Web Flow Conf is either going to make or break Web Flow. I hope to see a new vision for the future. As I'm taking courses and building client sites on Astro, it it's become painfully obvious to me that current web flow is a dinosaur down to even simple things like how component specific scripts are handled. I'm not going to be in the camp that uh fiercely defends an inferior solution just because it's the only solution I know how to use.

um all the WordPress people as a as a creator who has built my whole career around web flow. It's painful to see there are so many wonderful things about web flow that no other platform has which is why I hope so much that web flow can take a new direction. This came out right before that article. Not to mention that our video on Instatic also came out around the same time uh maybe a few days before this. And this this was a post by Timothy after he showed what you can do with sanity. Like this is like to us no code people. This is a whole new world and Web Flow has recognized that and they have admitted that they've been playing or they are playing catch-up.

So, is this whole post just us talking about catching up? Partially. Yes. Every serious tool in this category is racing toward the same conclusion for the same reason. Agents read and write code. So code has to has to become the real thing that gets everyone uh uh to the same starting line. It's not where we intend to stop. Now I'll go through the the article but the whole thing is about one simple thing is the fact that what you see in web flow is not the actual code. there is a intermediate layer that is breaking this whole thing that that is the reason why we have issues with the MCP. This is why AI cannot work with web flow m um like precisely and has so many issues because when you design in web flow you're not actually manipulating your website you're manipulating an internal representation of it proprietary data structure that describes your site which our systems then compile into the HTML CSS and JavaScript that browsers actually run the canvas you see is a rendering of that internal format.

The code that ships uh is generated output you never touch. This is very interesting because web flow has always been very close to the HTML CSS. Like the left panel is literally your HTML representation. Left is CSS but there are representations. They are not the actual layers. Like when you deal with a web flow slider, there are like other div blocks. when you deal with the CMS, there are like backend um backend code being run like my cat wants to get out so I have to very inconvenient. Yeah. What do you want? He wants to go chunky boy. Yeah. Chunky. He called you chunky. Yeah, he's he's a chunky boy. He wants to go out.

Okay, I'll I'll let you out. I think we need more cats in this show. I think I think Kabaza should be replaced with a cat actually because I'd much rather listen to a cat tell the news than Kabaza. Oh, hey man. How's it going? I have wireless headset. Damn it. Okay. So, so what is the issue with this? The issue is that some something gets lost in translation. That's issue. Uh, literally there is a compiler doing the translating. Uh, can you hear me? Can you talk? Yep, I can hear you. Ah, you were muted and I couldn't hear the noise and I was like, okay, it's fine. I'll continue. So, that's the issue. Web flow has a translation layer.

How are they going to solve this? any system with two representations of the truth joined by a translation layer will drift. So they they know the issue. The issue is like when you do this translation that the final result is not the same like you're not touching the actual code which we will get into. So it's a pretty good um article. I would highly recommend you reading it. It's it's written pretty fun. Uh anyway, so let's to let's continue to code sweet code as uh the canonical artifact uh the thing the canvas renders the thing agents manipulate the thing you own that's the direction web flow is going. So they are rebuilding Webflow around your code being the the actual code like the thing that you create in web flow being the final code and they would just show it to you.

So they wouldn't put a representation layer when you add a div block you would be adding an actual div block and that that's it. Um so what this means we will get to it uh or like how I think it will look like. Um, so what you make in web flow becomes yours in the fullest sense. Readable, portable, real. What portable is it like? Okay, th this is getting interesting. So hear this part clearly your site isn't go so okay before we I go there actually so portable this brings this creates so many questions around like what will be the value of web flow we will talk about this but before you get a panic what will happen to my website they have an answer to that so hear this part clear clearly your site isn't going anywhere and neither Neither uh is the way you work today.

I don't know what that why. What what is that? Neither is the way you work today. It is changing like what do you mean anyway? But your site is not going anywhere. We are building the bridge before we ask anyone to cross it. That I don't know like if they intentionally like picked these specific words, but this to me sounds like web flow 2.0. That's why I mentioned it. Like I feel like this is a new web flow. There will be the old web flow and they maybe build a translation layer to to migrate the old site to the to the new layer to the new tool. Maybe that that's my understanding uh after reading this and I hope that's the truth.

I I hope they won't keep the same architecture that they have. I guess they won't be able to do that for the most part, but I hope they won't keep the same CMS. The CMS that used to be amazing in 2019 today, as Timothy said, does feel like a dinosaur. Just touch any other CMS and you will notice that. Anyway, so they are playing catchup as I read. This is this is where I started. They are admitting that they are playing catchup. um the they understand that the developer wants to extend a component without fighting the canvas. That's the truth today. Like if you want to extend web flow, good luck. It's really difficult. The agent doesn't want an interface at all.

It wants an API and a boundary uh it isn't allowed to cross. So they want to create this in a way this sound to me like instatic what instatic already doing. So yeah I I I have so much hope for instic. So what is the next chapter creators um web flow recognizes that creators need an immediate connection to what they create for the web. What you create is code. So they are going as we said already they're going back to code. Um they are not this is they are not giving us any timelines. They are not giving us any designs. We don't know how it will look like. We don't know what is the architecture anything.

We don't know anything. They just announced this as a big announcement but like a direction that they are going back to the truth which is code. um and they will tell us more in uh September in web at at the web flow conf. Now I want to mention one thing that I uh on purpose left out for the ones who want to stay longer for the where is this argument coming from? It's a bit philosophical and I don't want to go too deep in it but I would highly recommend just reading this this article. It's very interesting. It's this video, Victor's argument. Uh, and I watched a little bit of it, but it it's amazing.

It's worth a watch. So, uh, Brett Victor, inventing on principle, and the idea is that what you create has to be like like on top of directly on top of like the technology. Like that's kind of like a lame way of me explaining it. Um it's just that the creator has to be in like direct touch with the medium that like runs it or creates it. Like it's that direct control. It's not having it's the opposite of having uh proprietary technology. It's the opposite of framer and Figma and frankly even web flow. Web flow was the closest to code, but it was never the code itself. Web flow was a translation layer. And for a long time, I personally felt like and I believe that web flow was actually like the final core, but it's not.

It it wasn't. It has never been. And now they're going back to that because of AI, because of agents, and because of creators to have direct access. So I would argue uh I would highly recommend watching that video. So a long uh a long explanation I guess uh what do you think like what do you how do you feel about this? So so many things and I just to say like you know instatic really is like that's why my money is on in static right because it's it's done what web flow needs to to do. It doesn't need to be open source. Web flow doesn't need to be open source. However, it's it's done where Web Flow it represents where Web Flow needs to be a really clean and clear way for agents to interact with um the website to be able to edit and do whatever it is that they want.

And uh the tool itself is not battling the agent. What they're basically saying in the article is that Web Flow and agents are fighting each other because it's not it the agent doesn't want a canvas. is very if if you can get somehow get that in the training data and I'll be releasing a video on this. I have used web flow MCP to a high degree of success with a workflow that I've developed that I'll yeah I'll I'll be making a video on as soon as I can possibly do it. Um there is a war going on here which is creating a disjointness. Right. With that all being said, Web Flow was always a tool that made it easy to create websites, right?

That was its goal regardless of that was its north star, right? Regardless of how it does it, that was its goal. Now there's an there's an easy and that is drag and drop. That was drag and drop user interface or whatever. It was really clear. anyone can do it and um that was why it blew up and that's why it gained a lot of traction over something like WordPress. Now there is an even easier way to create websites which is through agents. Web flow obviously serves many other purposes than just for the creation of a website. As we were saying before the show, many many website many many people that create websites literally design and build their website one and done and don't touch it for years, including myself, my own personal website.

I don't continually update it. This is not necessarily the correct or the best way to treat a website. It's an ongoing beast. It should be maintained. It should be uh I don't mean maintained as in security or anything like that. I mean it needs to evolve with the current climate. It needs to um respond to user behavior. It needs to respond to data and things like that. And then the design and the and the functionality of the website needs to change ongoing. That's that. But that's an ideal scenario. And you know, unless you're a big enterprise with loads of money, you're not really going to be doing that. So there is a place that Web Flow is is still really really good for um that does not need to change whatsoever and that is well I say it doesn't need to change obviously you said yourself the CMS there's some improvements that need to be made from the CMS but from a UIUX standpoint you create your collections you add your fields correct me if I'm wrong but that's a pretty nice system it's just the limitations on what you can do with those collections and the fields that it can you want complete flexibility.

You want to be able to have a component fields as an example and then just drag that in and that can just feed into whatever. Point is the the the if you break web flow up into two three/ird um is the actual build of the creation of the website. It's the visual manipulation of the website and it's the content of the website from marketers from marketing teams or whatever. twothirds of those are still pretty strong. I think Web Flow has it's just the creation that needs to be rethought and um I don't want them to put too much or I don't want us thinking Web Flow is this hole tool just because one of the twoirds is falling behind.

It's still a really useful and powerful tool. Again, I said this before the the show. We get so autistic in our own minds that we don't we can't we don't think about other people. We only think about how we interact with our website. Whereas, as actual fact, we're just one/ird. As builders, we're just oneird. We might be two if we're maintaining the design and we like the drag and drop and this and that. We might be two/3s. But there are and their business method their business uh shows that the that's that third third which is the marketers the the marketing teams that maintain the content and whatever are a massive part because they've just driven so hard into all of that in the last few years and and kind of less the creator community in the dust because we're just such a small part of it.

So I do have hope. I think they can remedy it. Um and they and web there is still a time and a place for something like web flow for many many businesses and organizations. It's just they need to rethink this creation aspect and in part the maintenance of the design aspect because people might still choose to use agents to be able to manipulate and update the website from that perspective. Um that's Oh, and just to finish on, but all that to be said, my money is still on Instatic. Yeah, me too. I wanted for many many people and static is there. If this is your first time listen like hearing in static, you like you you you need to check out our video on it.

It Yeah, man. I I don't know. I feel like they are going at such a speed. It's open source. I have a list of PRs to do, but I this is my first time doing PRs. Like I'm not a developer, but I'm like like I'm doing I'm I'm noticing bugs and I'm like helping like with the help of my agent, I'm trying to understand how these bugs like why do they exist? Like where do they exist? How can we fix them? Anyway, so I'm I'm a bit I haven't uh done a PR yet, but I want to do some. You've changed. I want to do some it. It's fun. Like why am I mentioning this?

Because like I can't do a PR with web flow. I can't like if there is something broken with it, I have to ask them and wait for them probably weeks and months to to solve it. Yes, to be fair like some bugs they they they fix it immediately to be fair. Uh it's nice that there's a support exist like a community exists but I see the same thing you know happening to in static why not like if there is an uh if there is a bug I can use my agent to fix that bug if I need a feature I use my agent to build that feature into the app the app is molding into is it molding like I'm shaping the the the app itself into what I And I think that is such a strong point that Web Flow probably will never have.

I don't know if they will expose the the back end, you know, the how the database works like we probably will never be able to. I don't know, maybe they will, but within static, I don't think they need to though. I don't I really don't think they need to. Like I think web flow needs to needs to hang on to the core aspect of why people choose web flow is that stability and security right they don't need to they don't need to and this goes back to the that we the builders who are winging and whining about that we need more from web flow they're just onethird I don't think they need to give up some of their strongest the strongest elements you still need some of these people like again with Jensen opening and signing this better of open source like you need both kinds of people you need you need that so so my question if I build a tool around um in static where I like manage and host in static for you like you the same way you press a new you know new site in web flow and you get a new new new site what if you build a wrapper around in static put it on you know best um hosting infrastructures ever like the same things that literally web flow uses Cloudflare right AWS what if you put it there and then you have hosting and there is no even installation you you handle all of that so the same way you would create a new web flow website you would create a new instatic and it would cost you at as like significantly less it would cost you a fraction of it.

So why would you go with web flow? But again like you're taking on the responsibility, right? If in static introduce a bug and it takes down all the websites or you you decide to change in static source code and you introduce a bug and all of the websites go down or some you know it plays nice with twothirds of the website but then a a very very small amount. But isn't it isn't it true about web flow? Like my clients come to me after they break their web flow website. Either themsel break it or they give access to somebody else, they break it. Yeah. Well, if if you're if you're happy to fix that, if you're happy to fix that, then yeah, sure.

But I'm not I'm not quite sure I'm seeing your argument because at the end of the day, like you people pay for convenience, right? Whether it's you or whether it's people pay for convenience. That's my point. I'm saying like for me it's it's not difficult to install in static but it might be difficult for other people to you know to do that. So convenience will be created for them. So if they if the barrier is installing and managing in static and hosting it I mean they already have a one-click deploy to uh railway but let's say somebody will build a feature around that a service around that that manage like managed in static the same way we have managed uh WordPress right you don't have to deal with hosting WordPress yourself there are plenty of tools that have built you know systems around it even like with plug-in updating systems.

I think my point is I feel like Web Flow doesn't have a lot to offer, but they still have like they have their UI, they have um I don't know. They have their hosting, but I don't see why would I not go with Cloudflare if that's you. You you you absolutely 100% do that if you're happy to take on the risk and there's not a tremendous amount of risk. I was doing air quotes for those on Instagram. Um, if you're happy to take on that risk, an element of like you are responsible for that app. Let's say your client comes to you and says, "Oh, I don't like the left the panel, right?" and um they want something changed or they want to be able to do something within the UI, not the website, the UI of Enstatic and you change that and then that because of the way you've built it affects everyone else's installation of Nstatic, then you're then you are then responsible.

You you're beholden to that. And don't get me wrong, it is you're right, it's no different to what Web Flow, it's no different. they still have to get they can still do this exact thing. I'm sure they've introduced bugs all the time and I probably remember a few right they are still responsible for that but people indeed pay and that's why there was so much anger um on a bug m maybe a few years ago where something went wrong and it was like but you you pay web flow a lot of money to make sure something like this doesn't happen you know but they're being fed by investors they're getting paid salaries you yourself might not be getting paid enough to really um would deal with that.

But all that we said, there are many many many people who still prefer WordPress who don't see any need to go to web flow because it's whatever cheaper. If you if you got if you're happy to manage a static instant instatic instance then you can do that but don't like just don't ignore the fact that people will feel anxious about little old Kabaza there solo dev who um just has just stopped using fable to start a server. They don't need to know any of that sort of stuff. But my point being they just might feel bit they might feel more um because because web flow will offer SLAs. Web flow will build contracts for their enterprise users.

It's not just a straight up uh what do they call it um offtheshelf service or anything like that? People will pay for it. Would you not be able to do the same thing for your enterprise clients if you let's say instatic is very early, right? Yeah, but let's say let's say Instatic gets because we're we're at the same time I don't want to have too much hope and then in static becoming something that I don't know they live leave um a month from now. We don't know that, right? It's it's open source and you've just keep downloading the latest version, man, because then you've got it forever. Yeah. Yeah. Um well, I'm doing that. I'm like updating it constantly because I want to see like what they are introducing and not.

But my my my I my point was I don't want to like like communicate that in any way that instatic is a complete solution. It there it's an alpha version but the alpha version looks like really in it looks incredible. By the way, now that we have talked about it so so much, let me do this. They have a website before we go. I think Yeah. Uh they have a website. They didn't have even a website. It was just GitHub. Um you were just looking through to GitHub. But nobody gets hold your website hostage. I I reckon they know I reckon they know who they're going after. So um yes, if you're asking me would I do it 100%.

I'm planning on recommending it. I have already recommended in static to a client cuz I am prepared to take on the risk air quotes uh of this service. Yeah. But that's not the point. The point is that there will be companies who want something more, who want some uh reassurance from someone slightly more um uh who has business insurance up to like 35 million like you you I got my teeth straightened, right? Yeah. I went with a service. It used to be I don't know whether you know this service called Invisalign. No, that Invisalign is these plastic um aligners they call them that progressively Mhm. shape the the the first one you put in your mouth fits almost identically to the teeth that are all crooked and whatever.

I see. You hold it in there for a week and then you change to the next one moves it even more. Slowly but surely your teeth become as beautiful as mine. our their patent ran out. Their patent ran out on their technology and then all of a sudden an influx of copycats came to the market and you could go and buy the exact same technology, exact same this and whatever for like a fraction of the price, right? I was like, I'm having a bit of that. I'm not going Invisalign. It's it's four times the price. I'm going to go with a cheaper version. Works exactly the same way. my friend's girlfriend um for some reason were like nah I just want to go Invisalign I just I'd rather pay the money and just go invisal line so I know I'm taken care of it's exactly the I could see that it's exactly the same thing right but some people just feel more confident with a reputable company who has a track record who can who can show the receipts on hundreds of thousands of people using their service it just you just need to escape this autistic mindset that we all have of that just because I want it, just because you want it, doesn't mean everyone else wants it.

Like, you just have to accept that. And I think Web Flow are going to be they're not going to they're not going to disappear anytime soon. They're going to have plenty of uh customers still very very happy with the service and they're still going to have many customers who never even touch MCP or an agent. they literally just drag and drop or start from a a theme and build their website and and they're more than happy with that. So all I'm saying I'm just trying not to be ignorant thinking just because I don't see the you I didn't see the use of an uh of web flow when um prime pinrow came out I was banging this drum of like yes but I want access to the code and even though the UI is ugly and you said this in our an instatic thing you said yes the UI is ugly but that doesn't matter because the job gets done and that was the line I used with Pinerow I was like pinrow looks ugly.

but it gets the job that I want to do done. And now they have an MCP and that that this all runs through agents now. But the problem is it doesn't have a CMS. It's a static website builder which many of the websites I was using were building were static websites. So this is this was fine. To be fair, this is a different level of ugly than uh in static. Instatic is partially ugly. No, but everyone th this line that you you drew this exists on different areas for for different people, right? And so my point my point is it's like you know the the people will just have different requirements and we shouldn't base our whole opinion just on our our own personal opinion because everyone's different.

Yeah. Well, it will be interesting to see where Web Flow goes. I on this and I I'm you bet I'm going to watch the web flow conf and I want them to succeed. I want them to be the best tool out there. It's the tool that I built my career around. It's the tool that I started with. It's the tool that you know we are a part of the community. I think they've already lost you though. I think you're in static. You literally don't see a reason not to use instatic. So they've already lost. Well, I don't know. Like may maybe I'm I need to play more with the instatic, but right now I'm doing three tests like actual like m one build and two migrations with instatic to see how it is.

I I need to see like in a bit more like long-term. I fully see your point with like bigger clients especially where it's more sensitive. Yes, you do want it. Even if it's the reputation, you know, if I tell a client, let's go with instatic, they might be like, what is that? But if they know web flow and there is the reputation and there is a support, I mean, I sent my clients to web flow support if they need it, right? Y if it's not something that I do, I send them to the web flow support and I'm sure there will be somebody to answer and I know they are great at it. They I know they know their stuff.

So there's like a lot of things like that uh that it's not a given. It's not a given and it's it's valuable still. Yeah. All right. Uh well, we'll see what happen and we will we will cover it. So subscribe for the Web Flow Con coverage. Right. Listen, we both got a hard stop, so I'm not going to dwell on this too much, but AI has introduced a huge problem. We're writing 10 times 100 times more code. Like it it's absolutely insane. So it's created this brand new problem of how we manage the flow of code eventually into production. And this is where GitHub finally I think this is an old feature. They finally have announced the support for stacked pull requests.

Does this make any sense to you? Stacked pull requests. Do you use git in this way? Uh well, I just told you like I'm literally starting to to uh like get more involved. So I I'm not sure. But like why would you need stack pull requests? So when you work on a a feature um and we we're we're so tempted to create feature after like a lot of features in sort of one session. Yes. you might have separate like pockets of code that depend on each other, right? Or that um you you want to maintain. Well, we'll get into it, but you want to review separately. Like you don't want to have to just review when you when you have a pull request from one branch into the next.

You don't want to have to look through all of that code or your specifically your agent doesn't want to have to look through all of that code because there's a hell of a lot of context for it to figure out what the hell is going on. So stacked pull requests allow all of your individual features or or um workflows to be kind of separated out into their own single pull request that all get just merged together and um or squished together and merged when they all pass and they've all been reviewed and and whatever. So uh where can I show you uh an example or a quick rundown? I mean, they're an ordered they're an ordered series of pull requests that each represent focused layers of changed.

With stacks, you can independently review and check each pull request, then merge every everything together in a single click. No more opening a single large pull request that takes forever to review or splitting work across multiple branches you have to keep track of manually. Because here's the other thing, that's the other thing about it. If you create separate branches or if your team is creating separate branches or whatever the dependencies between the two get lost and uh let's say you create a branch and as a feature and then someone else wants to depend or you you know need they need to pull down your branch and then create a branch off of that it just kind of gets messy.

So this is this is really nice. Um, I think that it will be really helpful to Let me see if there's anything in this this uh thing here. No, there isn't. Um, it would be really helpful just to break things up. So, here's an example I've got of a stacked pull request. And basically, this change depends on this change depends on this change or vice versa. I forget which way this round this goes, right? And it's just as some documentation that's been added and all the rest of this is just how it looks. And once all these review, these are all drafts. Ignore all of the uh little semantics aspects for this. These are all graphs.

If they're if they're um not, I would be able to merge this stack. But if I go into an individual one now, I have my boy Claude, uh, sorry, my boy Code Rabbit come in and review all of my pull requests and comment and, you know, suggest changes and this and that. And of course, going back to this idea of code reviews and stupid amounts of context, this can be a lot more focused and a lot more um impactful on this this specific pull request because it's just got it hasn't got loads of other bits and bobs that is muddying the context for it. So, you can just uh so it can review it independently and whatever.

And then all and then all of these kind of get uh stuck together and I can just merge all of the stacked pull requests all in one. So really good for the way we're writing code. Really really good for the way we're writing code and and um able to have a workflow that merges it. So pretty cool. Um you Sorry. Nice. Uh I will I will ask my agent and probably start. So you will Yeah. So you just got to you got to have the CLI installed. So once you've had the GitHub CLI installed, you you install the extension here which allows it to work. And I've forgotten where it is, maybe on Twitter somewhere, but you've got a GitHub skill that you can install.

Ah, so the skill understands how to create stacked uh thing. But I was going to say as well uh maybe if I just quickly unshare my screen. Uh stop and share screen here. Um, I asked Claude what would be I'm a junior developer and explain to me what uh how this in practice works. And it starts with you knowing that you're going to create a stacked pull request like and you're you're architecting that in your mind before you even write a line of code or your agent writes a line of code. So it's not just a case of you have a bunch of stuff and well you might be able to actually but you could say if you've got loads and loads of code changes ready to be pushed up to to Git.

It's not just a case of saying stack this pull request. However, it could probably figure it out. That would just be a complete waste of time though to do it. So it really does um yeah so you start with this mindset. You start with a mindset and how you're going to break it down and you then create one stack or ask your agent to create the start the stack and then you create a new stack that depends on the stack that you've or stack you've just done. So you just build and build and build until you're then ready to to push. So I won't read all this out here but the important aspect of it is is that you want it to go into this with the mindset of creating a stack something to play around with.

I think if you have code rabbit reviewing on GitHub, if you're really into like I do I literally use branches just so that I have a second pair of eyes uh that code rabbit reviews. Um it's it's really beneficial. I I I need to just use this in action, use this and and figure out how it really really works for me, especially as a solo dev. Like is this better for solo devs or is this better for teams? yet to be seen, but it literally just got released yesterday. So, yeah, pretty cool. Cool. Right, I've got to shoot. I'm already four minutes late. Um, that is all we've got time for. Like, subscribe, subscribe wherever you are.

Shout. How many people on uh We got five on YouTube. Five on YouTube. Yeah, hit the like button. Or hype. There's a new thing called hype. I don't really understand what it is. Um, tune in next week for all of the news in AI design and web. Um, I've been Sam and I've been covered. You know, I'm doing, you know, I paused there now because I know that you're going to jump in and uh, it's so funny. It's so funny. I love it. Right. Keep on vibing, peeps. Peace out.