EPISODE 1912

[0:00:12] GV: Hello, and welcome to SED News. As I'm sure many of you know, this is the different format of SE Daily where we cover tech headlines. We go into a bigger topic in the middle, and then we look at some Hacker News highlights towards the end. As usual, it's myself, Gregor Vand, and we've got Sean Falconer as well. 

[0:00:34] SF: Hey, everyone. Hey, Gregor. 

[0:00:35] GV: Hey. Yeah. So, we are recording this a little bit closer to our last one than maybe usual, but there's been no shortage of news. But as usual, yeah, I think we've also been doing a lot even in the last two to three weeks. So, yeah. What have you been up to since we last chatted, Sean? 

[0:00:51] SF: Personally, I guess the big news on my end is I bought a new house in San Francisco. 

[0:00:55] GV: Oh, nice. Sweet. 

[0:00:57] SF: I'm trying to do all the things that you need to do to buy a house here, move while also simultaneously trying to keep my work life going, and also my kids fed and showing up for school and stuff like that. So, it's been a lot to balance, but it's exciting. And then I have upcoming travel. I'm off to Seattle and then Washington DC in a few weeks as well. 

[0:01:17] GV: Yeah, busy. Well, again, great that you can also be on SED News. 

[0:01:23] SF: Absolutely. I mean, this is, of course, priority number one. Who cares about those kids eating or not? We got to get the news - 

[0:01:28] GV: Yeah, delay signing of the house for a day so we can get SED News done. 

[0:01:33] SF: Exactly. 

[0:01:33] GV: Yeah. On my side, I was actually in Tokyo all of last week. My work setup basically allows me to work from anywhere, which is very nice. But I mean, I do base myself in Singapore most of the year. But yeah, Tokyo, favorite city of mine. But yeah, I hadn't been in a couple of years. I was hoping to see maybe a little bit more tech advancements there. I guess what I mean by that is it still doesn't feel that when you're kind of using public services, technology has been thought about. It's always like an afterthought. 

For example, there's a high-speed train from the - well, Narita airport, which is the one that's further away, actually, from the city. So that's why they have a high-speed train. And yet to get the ticket for the train, even if you pre-purchase it and you have a QR code, you still have to go up to a desk with a human, and you still have that human scan and give you a physical paper ticket. Yeah. And there's maybe a way around it if you are Japanese and can speak Japanese, but to me, walking into what's supposed to be one of the most world's advanced cities, you can't even get on the train from the airport without going to a physical ticket desk. That's what I mean. Yeah. 

[0:02:46] SF: I mean, that seems surprising to me. Maybe somebody that's Japanese or speaks Japanese can let us know if there's a workaround to that. But it's been a long time since I've been in Japan, probably over 15 years. And I remember at the time when I was there, actually, it's almost closer to 20 years, I was blown away by essentially what they had as pre-smartphones. People watching television on their phones on the train at that time. And I'd never seen anything like that. I don't even know. I guess iPhone was just getting launched in the US at that point, but it was very much like I was a student. I wasn't walking around with an iPhone. That was for the wealthy population. And I'm kind of curious what has happened. 

[0:03:26] GV: Yeah. I mean, I have the same memories. Yeah. I mean, I didn't go to Japan when I was a kid. But seeing, yeah, videos of people watching TV on their watches and cars had like TVs in them way before they sort of came outside of Japan. But no, I mean otherwise things still do feel quite sort of advanced there otherwise. I mean, the systems that they have running generally are very good. The train system, despite what I just said about the tickets, the train system is very good. And we did that little piece a while ago on the IC cards that they use there. And yeah, I've still got mine. And very fast, very easy to use. 

They do still run a lot on cash as well, though. That's my other interesting - they've definitely moved on since 2 years ago. But there's still quite a lot of cash-only places. And I don't know. There's a very good system, I think, for buying food, buying ramen, which is the machine is actually outside the restaurant. And you put your cash in, and then you select the meal, and then it gives you a little ticket. And then you go into the restaurant, and you give your ticket to the chef. The chef doesn't have to handle the payments or anything. You've already kind of bought your meal. And I've always thought that's a really cool system. It's not like that technologically advanced. It's just like a vending machine that you put cash into and get a ticket out. But I think it's a very elegant way of solving that. And it stops the chef having to do anything, any administration. 

[0:04:48] SF: That's interesting. 

[0:04:49] GV: Final point, it was a great experience. Flew Japan Airlines, and there was a double-side of A4 instructions on how to use their Wi-Fi. And it just made me think about Starlink, where there are no instructions, you just connect to the Wi-Fi. So, I just love this. Someone had to put together double A4 pages for their Wi-Fi system. 

[0:05:08] SF: Yeah, that seems overly complicated. It's a little bit - I don't know. Have you had any experience trying to set up a printer recently? 

[0:05:15] GV: Not recently. But yeah, I can - 

[0:05:17] SF: It hasn't changed. It's the same if you did it 30 years ago. I mean, maybe there's Wi-Fi now, but it's still a pain. It's overly difficult to get a printer to just work. 

[0:05:27] GV: Yeah, there's a DX learning in there somewhere where it's just like two sides of A4 versus just connect. Let's hit the headlines. Unsurprisingly, we're going to be talking about OpenClaw. This has been obviously quite a whirlwind month for the founder, Peter Steinberger, an Austrian guy who had put together what is now known as OpenClaw. It was Clawdbot. 

[0:05:53] SF: Yeah, Clawdbot. But not spelled like Anthropic's Claude. 

[0:05:57] GV: No. But even so, Claude or Anthropic took issue with that, which is then it changed to Moldbots, I think. And then it's OpenClaw with a sort lobstery icon around that. But yeah, I guess the headline is that he's actually now joined OpenAI. You had Anthropic telling him cease and desist on the name. And then you have OpenAI saying, "Hey, come join us." So yeah, what did you make of that, Sean? 

[0:06:20] SF: Yeah. I mean, for anybody that maybe really has had their head in the sand for the last few weeks, this really blew up, I think, in the last probably like month or so. But in terms of what OpenClaw is, it's an open-source, self-hosted autonomous AI agent that runs on your local machine that you allow it to interact directly with your files, your applications, web services, whatever things that you want to grant it access to. You give it your API keys and so forth. 

And it acts like a 24/7 personal assistant. Uses LLMs and skills to kind of automate these different tasks for managing your calendar, sending messages, writing code. It has hooks into being able to react, have conversations over WhatsApp, or Teleport, or whatever kind of messaging interface you want. So, you could be on your phone and go tell OpenClaw to send an email or to go off and manage your calendar. 

I remember years ago when I was just a young engineer out there, there was a guy that I worked with that kind of proposed this idea of using email as a transport mechanism where you could send in almost a bash command over email. And then you would have something that would listen to that and invoke the bash command. So it's like, "Oh, I want to know what files are on my computer at home." And you would just send it over email. And the idea was everyone knows how to use email, so let's just do that. And I was like, that's a cool idea, but this seems extremely dangerous to do. We never ever did anything with it. But this is basically the agent version of that with crazy superpowers. 

So, at the same time, I think enough people obviously found value in it to actually install it and use it. And I think it created so much buzz. And I think part of it is almost like you can't turn away from like an accident or something like that. You're watching like a train wreck. It's doing a lot of things that everybody says not to do when it comes to AI or all the scary things. 

[0:08:11] GV: Just basically open this up to your local file system, basically. 

[0:08:15] SF: Yeah. Of course, security people freaked out about it. I'm sure, given your background, it was probably really scary. And then there was also I think some things where some of the API keys, people were able to sniff out other people's API keys. And then on top of that they launched Moltbook, which was a social media platform created specifically for AI agents to interact, and post, and comment. And essentially, they use the language of humans can only observe. And that I think also plays into a lot of people's fear around everything that's going on with AI. And there was a ton of stuff that caught the attention of everybody in the space, basically. 

[0:08:48] GV: There were sort of like articles on there, like my human has asked me to write this. Yeah. It's sort of watching a bunch of agents' computers talk to each other was kind of what that was meant to be. And as you say, it was sort of almost like knowing that that's going to really stir the pot a bit to have that kind of stuff. Yeah. 

[0:09:05] SF: Yeah. And then even Peter, who had had a successful company in the past and basically came out of retirement because of everything that was going in AI, and eventually led to this, was he came out, and he would say, "I push the prod without reviewing the code." Which again is another thing that people have a lot of fear over. 

I think all these things in combination really created a lot of fascination and discussion around it. But I think the outcome is incredible. I certainly didn't see this that he would sell this to OpenAI and now be at OpenAI. And between, I think, what he's getting cash-wise and stock, is he the first sort of one-person billion-dollar company? 


[0:09:45] GV: Yeah, that's interesting. We sort of ignored that topic for a while. But yeah, this could be that. Yeah, the push to prod without looking at it. Well, I can believe that because I should probably say, disclosure, I work at Supabase. He used Supabase for the back end, and did not turn on role-level security. 

As usual, Supabase got the rap for that. Why can we see a database? Blah-blah-blah. And as our usual line on that is turn on our RLS, it's very simple. Just turn it on. Yeah, pushing through the product without having reviewed anything. Yes, that's I guess what happens, unfortunately. 

But I mean great use of technology. I mean, I think we're going to get into why this over something like Claude or Claude Code, etc. But yeah, I think this has been a very interesting wake-up for like how fast things can still move. 

[0:10:36] SF: Yeah, I think one of the things that it did as well is it kind of gives a glimpse into the future of where things are probably going to go. Maybe in mass, people aren't going to use something like OpenClaw because of all the security concerns and so forth. But there is a subset of the population that found it valuable enough. Obviously, it created a lot of intrigue. But someone's got to kind of crack the code on how do you take something like this, creating a personal assistant for people that can kind of manage their calendar and do all this kind of stuff for them, and do it in like a safe way. That's going to probably come at some point. Someone's going to crack that. And maybe that's partly why OpenAI is so interested in acquiring this and the talent as well. And if you can do that, that's a huge business that goes well beyond even what we're seeing in agentic engineering today. 

[0:11:25] GV: Yeah, definitely. So, our next main headline does also relate to OpenAI, I guess, given OpenAI just sort of semi - I mean, just going back to OpenClaw for a second. Yeah, although Peter has been sort of acquired into OpenAI, yeah, they were at sort of pains to point out that OpenClaw will be managed through a separate foundation. Yeah, I think we're probably going to see a bit more of things like this, where it doesn't look good for OpenAI to say they've sort of acquired OpenClaw, which is open source. They have to, yeah, work around that. 

I mean, obviously, they're going to benefit, I imagine, from the underlying technology somehow as well. But if they want to move that fast, they have to, I think, usually separate the code, the technology from the person pretty fast. 

[0:12:11] SF: Yeah. 

[0:12:11] GV: Moving on to OpenAI again, but this is in the context of ads in chat or LLM chats. It's something that we'd kind of touched on, I think, a few SED News ago, like how is this all going to converge? If you looked at chatbots taking eyeball share away from Google, especially. And then you had Perplexity doing their browser thing, which I've not heard a lot about recently. So who knows where that's gone. Again, that was like a whole thing about eyeball share. 

And I think it's fair to say ChatGPT is still the most widely used. I mean, I don't think in maybe our circles. I feel Claude is - 

[0:12:52] SF: I think in the B2C space is probably the case. But I don't know the stats for sure. 

[0:12:56] GV: Yeah, exactly. But yeah, I feel like Claude maybe is seen as more the professional's choice, especially in tech these days. 

[0:13:02] SF: Yeah. And certainly, Gemini from Google is on the rise, too. And then it's just there if you have the Google products. 

[0:13:09] GV: Yeah. But I don't often hear people going, "Oh, ChatGPT blew me away," especially not in the work context. Yeah, it seems more like that's just the one that most people have in their head. So that's kind of where they go to. And yeah, they now seem to be rolling out ads within the platform. And I think the TL;DR of why is just money. I mean, of course, it's money, but it's like they are literally not running out of money exactly, but we've also covered this, that they've got all these commitments. And there were some funny things going on in the last couple of weeks, where Nvidia is not going to invest something into OpenAI that they originally said they were basically. And that then had a knock-on effect to Oracle because Oracle was going to be supplying the inference for that. 

And where this is going is that Oracle basically made a statement saying this doesn't affect us. And the fact that they even made a statement means it probably does affect them. It's basically the commentary there. Why am I talking about this? Because money at OpenAI is a huge question mark. And it seems like ads is where they're going to plug the gap. 

[0:14:11] SF: I mean, I think this is something that we probably predicted at some point, but not to pat ourselves on the back too much. I think it's kind of like an obvious thing that probably a lot of people were predicting for at least one of these chat LLM services to eventually explore. And if you're in the B2c space, classically, how do you monetize consumers, especially when the value that you're providing is really about eyeball capture? It could be a game. It can be social network. It can be search. All those things have been monetized largely through serving ads. So it makes sense to serve ads here. 

Now the big thing is, and I think we also talked about this, is does this signal the end of the heyday of LLM-powered chat? Because just like we're in sort of this utopia world with video streaming at one point when you could essentially go to Netflix years ago, and Netflix had everything. You could just stream it. It was very cheap. And then all the providers of that IP eventually woke up, and they're like, "Why are we giving our IP to Netflix? We should go and create our own service." And then you ended up with now everybody. You got Disney, you got Peacock. Everybody has a service. And to cover the breadth of it, you end up with like 10 of these things. And basically, we're back to like cable television, except it's like on demand. 

So, we're certainly out of the heyday of video streaming. And I wonder if this is the start of that, at least from a consumer's perspective. How do you get value out of the consumer? And I don't think charging consumers a monthly fee is probably something that necessarily works. It's like how do you control essentially the rate limit on that at a value of $10 a month or whatever? And token cost is going down. Certainly, speed of tokens and the infrastructure costs are going down. But at the same time, they also need to generate money to invest in new infrastructure to train the next generation model, which is very, very expensive to do. 

[0:16:03] GV: Yeah. Exactly. Yeah. I mean, on the subscription point, it's sort of been floated really that, behind the scenes, Anthropic at least, they make their money on - or they can balance the book, shall we say, on enterprise. Enterprise is paying like full whack or even overwhack, shall we say, for what they're using. And that basically subsidizes personal pro plans. A personal pro plan there's no way that a sort of $20 a month is what it's actually costing them to run that. That's the max, I think, that a consumer is probably willing to pay for if it's not linked directly to their employer, basically. 

[0:16:37] SF: Yeah. And I think one of the things I've heard, Dario, the CEO of Anthropic, talked about this recently, was that, at least in their world, they actually make money from a particular model. But where the cost comes from is the investment in the next model. While we're in this phase of really essentially an arms race between these different companies to build the next generation and the best models, there's continual essentially infrastructure investments that you have to make. 

And until all that stabilizes in some fashion or there's some other way of driving down costs there, you're going to always be in a situation where even if the current state of the model would make money in the long run, you're taking all that money and basically - and then some to reinvest in generating sort of the next generation. 

[0:17:25] GV: Yeah. And I think this is a huge open question at the moment is just like where does the bucketload of investment, where does that start to sort of plateau, so that it's a sustainable "business model" to produce? Because if you look at a similar version of this is just the chip manufacturing. Huge amount of investment and always needing to - like Moore's Law, always producing something that's faster, half the size, half the cost. But there were applications for that. There were very clear money-driven applications, phones, laptops, this boom in hardware, consumer hardware. 

Yeah, I think it's just where's the constant renewal? A consumer is not paying what a lot of people would pay, practically, per year for a new iPhone for a while. They're not paying that to any of these providers at the personal level. Yeah, it's kind of interesting. And if we look at the mechanics of these ads that are now coming in, ChatGPT charges USD60 per thousand ad impressions. Are you familiar with how does that line up with how Google used to charge, or have you got any - 

[0:18:33] SF: I couldn't tell you really what they're - I mean, they charge typically on you set a bunch of stuff around like budget, CPC, your bidding and stuff like that. Some of it's kind of hidden from you because there is like a bidding thing that goes on. And certain words cost more, and so forth. But I don't know. I wouldn't be able to map that to what ChatGPT is doing here. I did read that this sort of on par with, I think, what Netflix had launched for ads. And it's more expensive than Meta, obviously. 

[0:19:00] GV: Yeah, interesting. The ad space is just simply an area I've never been part of. So, I'm not sure either on the pure financials of that, and don't want to just regurgitate somebody else's numbers. Yeah, that's interesting. Interesting, 60 US for a thousand impressions. So if that's, yeah, sort of similar to Netflix. That's interesting. 

[0:19:15] SF: I mean, another move that OpenAI is trying to do, they're certainly trying to get into the B2B space more as well. And they've made a lot of investments there. And you might be able to get to a place from a business perspective where you can kind of subsidize the consumer model a little bit with the idea that if you have a lot of consumers using this, this also becomes the thing that they want to use in business, where the business is willing to spend more money. I think Zoom was very successful with that model, for example. 

[0:19:43] GV: Yeah. Yeah, that's a good reference. And yeah, a colleague of mine alerted me to the Anthropic anti-ads that were running. So these were not ads within Claude, but TV ads, and I believe Super Bowl ads. 

[0:19:55] SF: Old school ads. 

[0:19:56] GV: Yeah, old school. Super Bowl. And sort of the format is like you're talking to your mother, and then suddenly an ad for sort of do you want to go on a date with an older woman pops up. They're obviously taking kind of the extreme end of where ads could go. But I think it paints exactly the picture that they wanted to paint. And as usual, it kind of always gives - Anthropic and Claude just always have this, I think, cooler edge to the product. And sort of, "Oh, we're in tune with the people," sort of vibe, versus the sort of Sam Altman OpenAI, like, "We will tell you what you will use." 

[0:20:29] SF: Yeah, there's definitely feels like a cultural difference between companies. And I think to me, and I wrote a post on LinkedIn about this a couple weeks ago, some ways Anthropic kind of reminds me of like the early days of Google, where they're collecting a lot of - I mean, OpenAI is doing this as well. But I think Anthropic seems - a lot of very, very gifted smart people are drawn to that company, maybe, because of their culture or whatever it is. And they're really collecting a lot of this talent and this talent pool and kind of giving them somewhat of an unstructured environment akin to Google in the early days, where it's not like overly processed. I think it's a relatively flat hierarchy. And giving them the freedom to kind of just like move fast and experiment. And they seem to be thriving as a result of whatever it is. 

How far can they scale that? I don't know. I mean that was something that eventually Google had to put a little bit more rigid control over and add more processes. But right now, it certainly seems they're making a lot of smart decisions in terms of the investments they make from a product perspective. Sort of their stance on trying around what's important from models and how they think about even security in this space. 

[0:21:39] GV: Yeah. And then just a sort of sidebar off this, what's happening over maybe in sort of this side of the world, in Asia. And obviously, China has quite a few chatbots. And it is Lunar New Year as we're recording. Year of the horse. And not just the horse, but the fire horse. And completely separate topic. It seems to have swept the internet this year. I keep hearing or having people say happy year of the fire horse, who would never have talked to me about Lunar New Year before. I think it's like a TikTok thing. 

But Alibaba, they've got their Qwen chatbot. And obviously, Qwen models as well. But they decided to commit $430 million to offering free bubble tea and other items to users who ordered food through the Qwen chatbot. And that has actually driven more than 120 million orders through the six days. I mean, that's a different way to approach. I mean, you can't - I don't know. Maybe we'll see something similar where one of the providers says like, "Hey, if you use us to do a thing, you will get a physical good in return." But yeah, it feels like a slightly more - I see campaigns like this in Asia a lot more than I maybe do, say, in Europe or the US. But yeah. 

[0:22:47] SF: I definitely think agentic commerce is where we're going with a lot of this stuff. People already use these tools to go and ask for advice on purchases that they're making. And then they go to some other place to go and make that purchase. But it doesn't take that much foresight to figure out, "Okay, well, if we just connect the lines a little bit here." And there's been investments from Google in terms of standards around doing secure purchases through agents and so forth. I think that's not that far away, where this is sort of the default. 

[0:23:20] GV: Yeah, this is an interesting one because, yeah, as this campaign was around, yeah, ordering food. If you sort of play that through, okay, a user maybe hasn't realized or just hasn't tried to order food through the chatbot. And so yeah, they offer this incentive. And then once the user's done it once, they're like, "Oh, this is possible." And you assume that maybe half at least are like, "Oh, this is actually nicer than going through my food apps." 

I mean, I can definitely see a world having just like stayed in Tokyo, stayed in a hotel. My wife and I call it hotel hell when we basically end up on hotels.com. And two hours later, one of us is still on it. Are you making any progress? And it was interesting. I actually just quickly tried ChatGPT today and said like tell me some hotels in Tokyo. Because I was almost trying to see if it was going to give me ads. But it did actually recommend - one of the hotels it recommended was the one I stayed in. So, I don't know. That kind of got me thinking maybe this is where I should be going in the future to avoid hotel hell. But obviously, the whole booking process, I don't know. I'm still bit old school. I feel I just want to see the real actual booking. 

Moving on to our final headline, we have Mistral. So, the French model provider, but has Mistral Compute as well. But they've bought a company. I think you pronounce it Koyeb. And basically, Koyeb provides inference. And I think there's kind of two sides to the story here. One is that Koyeb provides inference and can bolster the compute side of the Mistral models compute. But then Koyeb also provides spinning up databases or containerized apps, and this kind of thing. 

As I understand it, behind the scenes, that's actually Neon. This is not Koyeb's own infra providing that. But I think it looks like that Mistral still sees huge value in that even if it's based upon you're two steps away from who's actually providing the infra. But we're seeing this a lot, I would say. We're seeing a lot now - and this is maybe a topic for maybe next time, Sean. But where do people like Lovable go in the future if we're seeing people like Mistral probably move in this direction, where from the model you could spin up infra, and i.e., end-to-end, create an app. I think we're going to be seeing a little bit of a step shift this year across all the players, I think, actually. Yeah. 

[0:25:43] SF: Yeah. I mean, I think that it seems like all the model providers are making steps to create more of a moat around their services by owning more of sort of the end-to-end pipeline. Making more of these things just sort of part of the model experience. Because if you just have a model, then it's kind of like, "Well, the switching cost is low. I can go with the best model or the one that's like the best balancing cost for whatever my workload happens to be." But if suddenly the model also has all these other things that maybe it holds on to important contextual data for my business or for my personally, or it makes it really easy for me to just like spin up agents and run them in cloud infrastructure, and I don't have to go use some other application to do that, then why go somewhere else, right? You're creating essentially one experience that kind of serves your end-to-end purposes. 

[0:26:33] GV: Yeah, exactly. Yeah, very kind of interesting. This is probably the first, I guess, pure acquisition we've seen in this space so far. A model provider outright buying the infra provider, if you like. I mean, Lovable. We reported on Lovable buying - the name evades me now. But Lovable, being Swedish, they also bought a Swedish - Molnett. That's the company. They bought Molnett, who smaller outfit, three guys there. Whereas Koyeb, I think, is 13, and just has clearly bigger customers and more sort of advanced offerings. Less of an acquire here, I think, and more of a full roll-out team and infrastructure, which is super interesting. 

The big topic, I mean, it kind of rolls off some of the things we've been talking about already. But we've just seen such an explosion of developments and advancements in the last four to five months in agentic engineering, really. Obviously, OpenClaw has kind of brought things to a huge head on that one. But if we look at Anthropic's Opus 4.6, and we've got GPT Codex, basically, it's even coming back into mainstream media that coding is being disrupted, which I found quite amusing. Because it's like why, suddenly, are all these mainstream media - the reference I was going to bring out was just my dad literally messaged me and said, "Have you seen this?" And I'm like, "Why? Why is my dad, who retired, not in tech, why is he suddenly reading things that are telling him that like coding is being disrupted?" 

I always look for these kind of bellwethers or canary in the coal mine kind of thing. What is the thing that's suddenly making people wake up to this? And it does seem that agentic coding is why, basically. It's showing that it's not just ask code to be produced and code comes back. But suddenly, all these other tasks are being completed in the background that is suddenly taking over developers' workflow completely, basically. Yeah. 

[0:28:34] SF: Yeah. I mean, I think that there's probably a couple things that have gone into this. But it definitely feels like in the last 4 months, what the developer experience is has changed drastically. And I think that it's gotten to the point where the tools plus the models are so good that even the naysayers are starting to realize that this is undeniable what you can do with these things. 

Because when you were doing kind of one-shot, you send a prompt to a model, and you're like, "Hey, generate this function for me, or take this function and help me like optimize it and stuff," that's helpful. But I think you're seeing 20%, 30%, maybe, efficiency improvements there or something like that. Whereas now with the agentic engineering using things like Claude Code or using Cursor, running Codex 5.3 from OpenAI, or whatever it is, you're able to take stuff that would typically might take you hours to do and do it one-shot prompt, and they're just able to do it. It's really incredible having spent some time using these tools myself in the last few weeks. I think, purely, there is really a transformational shift that's happening. 

And then on top of that, we went from sort of copilots in assistance, where it's really like autocomplete, your kind of one-shot, to now agentic engineering. And I think the evolution of that is now these Ralph Wiggum loops, and Gastown, and sort of multi-agent orchestration. And there's some debate over how ready those are. But people who are really on the bleeding edge are doing that. 

And to use Steve Yegge's language, it's like agents really take chat and basically put chat in the loop. And then these things like Ralph Wiggum, Gastown, and so forth are putting kind of agents-in-the-loop where you have multiple agents running. And you see these people talking about their setups online where they got six Mac minis running on their desk all the time, so they can spin up more instances of Claude Code so they can always have their agent running. 

And so I think that the combination of the model and the tools, getting to the point where they're really good at managing contacts. And then probably the explosion of the number of MCP servers that are available, and things becoming more easier for agents to sort of navigate. Clearly, to me, there's just this transformational shift that's happening. And I think that puts the entire industry in a place where there's probably some uncomfortable questions we need to be kind of asking ourselves. 

[0:31:07] GV: Exactly. I mean, just to kind of, I guess, semi-summarize to this point, the fact that models can use tools at all is already sort of quite a big step change. 

[0:31:17] SF: And use them reliably. 

[0:31:19] GV: Yeah. Exactly. And use them reliably. I mean, it still depends kind of on your interface to that. I've been pretty impressed with, actually, Notion AI recently, because it just has the tooling and the access to files, and Slack, and all this kind of stuff. I mean, when we say models can use tools, the service that is built upon a model effectively can use the tools. You've then got multi-agent orchestration. So, the fact that subagents working in parallel. 

And as you were kind of alluding to, we've got these frameworks like Gastown now, where you have kind of like a lead agent effectively. And I love Gastown's way of like a mayor, and you've got a - which other people do they have? They have - well, yeah, the mayor, the town, rigs, crew members. I mean, I think it's really cool. Just a great way to like add a way for sort of people to kind of get their heads around what's going on. 

And then something I've noticed just without any thinking of this previously, but this idea of like context compaction, where AI can actually summarize - basically, it's on memory. And then can keep working on things without hitting limits. We're beyond the stage, at least I feel, on any paid plan now. We're on the stage now where you don't really very seldom hit like, "Oh, I've run the course of my conversation." It's like you can always add to a chat that you've had previously, and it's able to do things with that. 

And then you start up, say, a new chat within, in my case, Claude. And I did get a bit of a like, "Ooh." That's a scary moment when it's like I'm asking something, and it's like, "And so based on this project to do with X that you're already working on." I was like, "Oh, no. It has all the context about me." I mean, it was good, but it was just a slightly strange moment when I was like, "Ah, it does reference back to these other things I've been working on. Okay, understood." 

[0:33:07] SF: Yeah, I think one of the big things that we get out of this is - and maybe we're not 100% there, but I feel like it's certainly going in that way, is we're really compressing this time from idea. From idea to at least POC. And we'll get to the point where it's like you're really compressing the timeline from idea to production as well, where essentially the cost of experimenting, it becomes almost zero. Other than the token cost. 

But I can just tell Claude to go do something. And one shot or maybe a couple shots, it's actually able to accomplish that thing. And I think the interesting thing from my perspective is if we are really taking this thing that historically has been an expensive part of being able to create a product is the time it takes to write the actual code. And you could press that cycle. 

Now some of that cost shifts to testing, and test harness, and stuff. And I think over time, maybe that will also start to go away is the LLMs are actually able to take over a lot of that work. What does that do to an organization? How do we have to think about - what is the modern organizational structure? 

We've been kind of using roughly the same kind of organizational structure for like engineering, product teams, TPMs, and stuff like that for probably 20 years. But in this world, where you don't necessarily need as deeper expertise in particular, like language or framework, and you can experiment more freely, do we need the same sort of ratio of PM to engineers that we've had previously? Do the lines between different technical roles start to blur drastically? Do you have frontend and backend? Do you have people who are expertise? 

I think, in many ways, the deep expertise in particular technologies starts to lose some value when you can rely on the agent to do that. Then it's like, "Oh. Well, you need sort of more broad expertise to understand architecturally how these things fit together." There's probably other things that come to play in terms of what the value of a person starts to transform into being, the taste. How many ideas they can come up with? What is that new choke point for a company if it's not sort of the laborious task of writing the code? 

[0:35:18] GV: Yeah, exactly. And yeah, I think you mentioned in passing around, are we seeing the end of the 10x engineer, basically? The 10x asshole, effectively. 

[0:35:30] SF: Yeah. I mean, we've all probably had you worked on teams or been in companies where there's someone who's kind of like the jerky engineer or jerky expert, but people tolerate it because they have deep expertise in something. Maybe they've been there for a long time, and they're the only person who understands the full breadth of this codebase. Or they really know the specific type of technology or something, and people kind of put up with that. But if the models are really good at that, does the 10x asshole essentially, does that role have a place in the modern AI-forward company versus other types of skills that are maybe higher value, like interpersonal skills, driving alignment? Sort of the human component of it.

[0:36:10] GV: I mean, I think looking at the actual tooling that has changed. And we obviously talked quite a lot at the beginning about OpenClaw. But partly, I think the reason it's really then also jolted everyone, first of all, it's open source. And it's not tied to one of the models specifically. And it's come out of, as we call it, effectively one guy. Okay, very talented guy. But one guy has been able to kind of put this thing together. 

And then I think for whatever reason, it's shown a lot of people what they can actually do with agents that for some reason Claude and OpenAI weren't able to fully show or maybe had sort of lulled people into a sense of, "Well, no. Claude and CHPT are 4X." And that's "nothing to do with agents", which is obviously they're trying to shift the narrative on that. But we've only kind of maybe seen that narrative shift very, very recently with Opus and Codex, basically. Is that kind of what you see as well? 

[0:37:08] SF: I think that's fair. And I think there's also been now this shift that's happened, where sort of the CLI and the desktop is becoming like the housing unit for the agent. And then the agent has the ability to kind of reach out to these different systems, whether it's like local systems, or could be things on the web and doing search, or talking over MCP, or whatever happens to be. 

And I think that's an interesting shift. I don't know that that's something that would have been obvious. And will it stay or not? I don't know. But it seems like people are almost more comfortable running these things within their local environment. And part of that probably has to be probably related to the fact that so much of this is kind of tied specifically to agentic engineering. 

And even though so many things exist on the web, a lot of engineers still prefer to run an IDE locally and have the code available to them locally. And I think Claude and some of these agentic engineering tools have done a good job of mapping to that existing way of working that engineers are used to, where you can do these things within your local environment while still giving them sort of the power to touch all these different systems. 

[0:38:16] GV: I mean, yeah, I think as you're kind of touching on in terms of where is engineering going. I mean, we almost touch on this at some point every month, I think. I feel we almost have to because there are just always these leaps each month, and something is being questioned around what is it to be a software engineer now. But yeah, I mean, I do see that the jobs that are sticking around are the ones where someone has, say, like broad knowledge over something. And it does require human judgment, interpersonal skills. 

I think the interpersonal skills is an interesting one because there's still a lot that goes on in companies, where, quite frankly, it is just people are not agreeing on something. And actually, often it requires maybe a third person to come and kind of be the person to get the two experts to agree on how you're going to get a path forwards. And at least I don't see almost flipping a coin. I don't see that two experts are going to go, "Oh, we'll just let OpenClaw decide for us," or something. Because then they'll say, "Oh. Well, how are you prompting? Or how are your agents set up? And so how can we trust this?" 

So long as a human is in the loop somewhere, which they very much are, there's always still going to be room for humans with engineering knowledge. But I think, yeah, perhaps more on the softer stuff, which is just a higher-level judgment or negotiation, effectively. Negotiation being actually a big one as well. 

[0:39:40] SF: Yeah. I mean, I think that engineering certainly isn't just about writing code. It's really about solving problems and operating software systems, and potentially operating those things at a massive scale. And I don't think that just having an agent that can code for you necessarily takes away all those problems. 

I think the big question that people are sort of having right now is do you need as many engineers doing this? If you take sort of the coding piece of that off the plate, can you have one engineer that's now able to do what 10 people did before? I think that's kind of the big sort of uncomfortable question that everybody in the world of technology is kind of wrestling with. 

And it's just if you see past trends, like we had the industrial revolution, that made the labor involved with things like farming not as intense. So you didn't need as many farmers. It's not like farmers went away completely, but you could have essentially less of them doing more work. And that led to an explosion in growth of humans around the world and all kinds of other things. There was like a ton of good things that came out of that. But obviously, there's like this transitional period that's kind of uncomfortable and not great for everybody. 

[0:40:52] GV: Yeah, absolutely. Yeah, I mean, I think we'll maybe kind of leave it there this week with the sort of main topic. I mean, it was just, as we were saying, looking at what has happened in the last three, four, five months that suddenly has caused the world to sort of declare that software engineering is over. Within our own circles, we've already been posing that sort of slightly existential question of like, "Who are we now? And where do we bring value?" But we can find the pockets that make sense. 

It's, I think, when the sort of "world declares it", and stocks of SaaS companies are going down significantly because the reasoning was, "Oh well, now that you have things like Codex, who needs a SaaS company?" And I just found that such a classic kind of financers' way of looking at the world. It's like, "Oh, we've read this piece of news about Codex. And suddenly, a SaaS company is no value in it anymore." I mean, yeah, it's such a knee-jerk reaction. Yeah. 

[0:41:47] SF: Yeah. I mean, there's so much more that goes into those SaaS companies than even just the code. The other thing, too, that I think is interesting about this is in some ways, I think, that these tools bring back the hobbyist developer. Because now with less time, you can create sort of one-off compelling app. Almost like applications are ephemeral. You can create sort of just try something, it doesn't work. Or maybe you just create it to solve some specific problem. You could spin that up without a lot of cost and effort now. 

It's kind of like in the 80s, there was less engineers, but a lot of people were hobbyists. They were interested in technology, and they were building things. And you could actually build software that eventually sometimes got used by lots of people, or games that got used by lots of people as a single individual. And I think we're now in a similar place. OpenClaw is a good example, where he was able to create - basically, one person create this system that was used by a bunch of different people. Got a ton of attention. I think we'll see similar things start to come up more and more. 

There's also going to be more garbage and more noise as well. So, how do you break out the noise? But I think that's one of the really powerful things. And I've certainly been doing a lot more of that kind of one-off projects, things for my kids, all kinds of stuff, because the cost of actually making that thing happen is so low. 

[0:43:09] GV: Yeah, for sure. And just kind of - I think it got to a stage, at least I found this, where if I was creating something more as a hobbyist, the problem being that the level of like UI, etc., is that if you want to show it to anyone, then just the expected level is always just increasingly higher and friction barriers as low as possible, etc. 

Obviously, just to plug Supabase. Things Supabase speeding up things getting a database up and running, and some auth, and that kind of stuff, so much was still slow, slow, slow before this came along. And you'd be still be presenting a very crappy looking, even just Tailwind, etc. Still, someone would say, "Yeah, but that's just Tailwind. Why doesn't it look nicer?" This is just a sort of hobbyist app. Yeah, it's interesting the time decrease to produce something that people take very seriously now, which is interesting. 

Moving on to our final, usually best part of the show, Hacker News highlights. Yeah, last time, we only really had time for one. So we wanted to make sure we got a few more in this week. Yeah, I'll let you kick off, Sean. What's the first one that you wanted to highlight? 

[0:44:20] SF: Yeah, so the first one that I pulled was this article called Old School Visual Effects: The Cloud Tank. I thought it was interesting. He has a bunch of articles and blog posts about sort of movie special effects. And he rants a little bit about computer-generated effects. 

And this particular post talks about how Cloud Tanks work, which I'd never heard of before. But essentially, they're these large water tanks that used to be used to create atmospheric effects. So, Close Encounters, Indiana Jones, like a lot of movies that were shot back in the 80s by Steven Spielberg and stuff. He used these big water tanks to create certain visual effects. And it goes into a lot of detail about how those things work and how they were constructed for these particular movies. A really cool post. 

[0:45:04] GV: Oh, awesome. Cool. I'll check that out. I assume we're talking sort of a Cloud Tank, if you want to be showing like, "Oh, everything's very misty in the jungle or something." I don't know. Is that kind of the idea? 

[0:45:15] SF: Yeah, exactly. Close Encounters, they used it to, I think, create some of the atmospheric effects around when the aliens were arriving on Earth and things like that. So, yeah. 

[0:45:25] GV: Oh, awesome. Yeah, I love scenes like that. So, cool. Okay. On my side, first one, it was called I converted 2D conventional flight tracking into 3D. It's user Kweonit. Thank you for posting that. So, I'm already a FlightRadar 24 subscriber. I can't quite remember all the extra things you get when you subscribe at least up one level. But I live on a flight path in Singapore, so I find it really fun to just look out the window and be able to kind of see what flight that is that's coming in. 

I actually live on two flight paths. One is a military flight path, and one is a civilian. So, it's kind of fun. You got a fighter jet coming in quite close to my building, actually. And then you've got like a civilian just slightly further away. And even when I'm in Scotland, again, there are planes going from Asia over to Europe because they kind of carve that arc over the top of the - yeah, this is someone who has kind of put this into more of a 3D kind of model, which is kind of cool. 

I just love when it's nothing like massively revolutionary, but at the same time, no one else has done it to my understanding, where you're actually just seeing a slightly nicer way of representing these planes. And they kind of show roughly the trajectory and the direction they're taking. 

I think, also, it helps with - you can pan, tilt, rotate, all that stuff. So you can actually see the altitude levels more clearly, which I think is also interesting because I think many people don't realize you will quite often have three planes stacked, actually, on top of each other, basically, on the same flight path, but 2,000 ft apart. 2,000 ft maybe. Yeah, I probably - No. Yeah, about 2,000 ft. 

If you look on some of the key flight paths, you've got like, yeah, three 777s like on top of each other. And it's true, but it's just that they have to take different altitudes. But these busy flight paths, that's just how they operate. So pretty cool. 

[0:47:20] SF: Yes, very cool. Very cool. 

[0:47:22] GV: Yeah. Also, you can just plug in whichever airport's nearest to you and it sort of zooms over there, but ultimately it's a globe thing, I think. So thanks to the people that put that together. What was your next one, Sean? 

[0:47:34] SF: Yeah. So this one's pretty cool. It's a guy reverse engineered Sid Meier's Railroad Tycoon for DOS from 1990. And he has a write-up about it. But essentially, he hit the memory, the game money limit, which I forget. It's some small integer value you can imagine from like a DOSs from 1990. But his kids also, they just wanted to be able to build railroads. So, he wanted to also make it like simpler for them to use it. 

He did a full ground-up port of the game and dug through all the assembly and rebuilt it to run natively with modern graphics. I think one of the other things he wanted to do was like the original aspect ratio is like 320x240 or something like that. So he wanted to make it - be able to be bigger. 

The article goes into a lot of these things that he discovered from like 1990s engineering, which you had to do so much optimization because you only had so many CPU cycles. You only had so much memory cycles or so much memory stuff like that. The game uses overlays that overwrite code at runtime. There's a bunch of frame buffer color tricks going on in there. He ended up fixing the money overflow bug. He made the rendering resolution independent. He removed some of the save limits, and he added a play alone mode all to it. Really cool feat of reverse engineering of this game from the early 90s. 

[0:48:55] GV: Nice. Has he been able to release the code for that, or does he - 

[0:49:00] SF: That's a good question. I didn't see that in the write-up I saw about. 

[0:49:03] GV: Yeah. Because I'm just wondering, you probably maybe can't just for like copyright reasons and stuff. 

[0:49:08] SF: Yeah. Well, I don't know. I mean, at this point, it's been what, 37 years? So, do the copyright laws still even apply? 

[0:49:15] GV: I feel video games is the one that there's always some weird thing going on, where someone somewhere has the rights to it and won't respond or just says no. And Golden Eye being the obvious one. Golden Eye from the Nintendo 64 cannot be re-released basically because three different companies technically own the rights to it, and they won't agree on anything. Never happened basically. 

[0:49:38] SF: Mm-hmm. 

[0:49:39] GV: Cool. Yeah, very cool. Almost feels like such an over-engineering just to fix a money bug. But I mean, that's great. 

[0:49:45] SF: Yeah. Yeah. I guess he really, really liked Railroad Tycoon. 

[0:49:50] GV: Yeah. My final one was I just actually - you just got very - a lot of points up over a thousand GrapheneOS. Just, yeah, kind of plugging GrapheneOS, I guess, which is kind of cool. Currently, the only disappointing thing is it's only for Pixel phones right now. I run a nothing - 

[0:50:04] SF: That works for me. 

[0:50:06] GV: Yeah. Yeah. So it's an open-source OS. Effectively, instead of what you would get as your base Android. It is based on Android on the Android Open Source Project. But it differs quite a lot, apparently. And the big piece is sort of security and much less tracking, effectively. 

Android, these days, is very just baked with so much Google tracking in it. And so they say it's distinguished by a hardening of the kernel and key components, which does make it less vulnerable to hacking. But actually, also, that has Google Play Services running in an isolated environment. So that sort of allows - you can use popular applications without having to grant the broad permissions that they mostly ask for, and it's hard to reject. 

Permissions, at least on Android, for me, they've got better. Some will be auto disabled if they're not being used, and that kind of thing. But I think there's probably a lot more going on than we realize. That's like why GrapheneOS does kind of exist. Yeah, just thought that was kind of cool. Great to see still innovation in the OS space, and especially on mobile devices as well. 

[0:51:15] SF: Yeah. Very cool. 

[0:51:16] GV: Yeah. So we're almost at time, I guess. But yeah, I don't know, looking ahead. Predictions. We always come back the next month and say why didn't we think about how this could have played out? But yeah, if you had to pick something that you think we might be talking about next month, Sean, what would that be? 

[0:51:32] SF: I think there's going to be more sort of verticalized versions of things like Claude Code. They already launched Co-work for Anthropic. I think they did some sort of financial services, like Claude for financial services. I'm just using Anthropic as an example. But I think that a lot of these companies are looking for other ways, I think, to bring this technology to the masses beyond just engineering. 

I think the tricky part, especially with some of the agent stuff, is that engineering is set up well, I think, for these environments to be successful because there are so many things where it's kind of like a harsh environment. There's ways of checking failure conditions. And we also have a lot of things set up to help undo actions. If you commit something and you don't like it, well, you can branch on git and then you can try something out in an isolated environment and so forth. Hard to do that in, I don't know, legal profession, for example. Some of these things is like how are you going to translate those things over. But I think that's where we're going to get. I think we're going to have sort of more deep verticalization of some of this technology. 

[0:52:37] GV: Yeah. I mean, I think maybe mine's sort of similar. But yeah, I think just based on what we saw with the Koyeb acquisition. I think things are moving pretty fast here. And I think we're going to see, yeah, someone, one of OpenAI, Anthropic. I'm trying to think who else it could be. But yeah, come out with something more towards this idea of if you're using this for code, don't run that code somewhere else. Basically, run it related to us. Again, trying to take on some of the "vibe code operators". 

I was talking to someone in DevRel last week, and they were convinced that actually a lot of the vibe code crew had sort of moved to Cursor, interestingly. Lovable was being seen as like not pro enough or. And sort of cursor was actually where they were moving to. I can't substantiate that. I just think it's interesting that someone in DevRel, that's what they were seeing. But that would suggest that everyone's moving a little bit upstream in terms of someone that might have started with Lovable 6 months ago actually is like, "Oh, that's not enough for me now. I want to take something else." Yeah, that would kind of feed into this idea of does Claude Code spin up a backend for you or something? 

[0:53:51] SF: Yeah, makes sense. 

[0:53:53] GV: Cool. So yeah, we will catch up again next month. Yeah, I hope the house - are you moving this month, next month as well, or - Yeah. 

[0:54:01] SF: Well, we'll start the move and then be in there probably the first week of March. 

[0:54:04] GV: Yeah. Nice. All right. Well, that'll be the first one from first episode, I guess, we do from that place. So, cool. See you there. Thanks, everyone, for joining. We'll catch you next time. 

[0:54:13] SF: Thanks, everyone.

[END]