0:00Let's get right to it. Um, you said that next year we will probably produce more software than the whole history of computing. So for developers sitting here, uh, is that a good news or a bad news? Well, first let me back up the statement because I think one thing that happens with
0:25Silicon Valley CEOs is that you can say anything and everyone gets a free pass because AI AI and I was recently looking at our own data of the evolution of Forcell. For those that don't know, I started forcell in November of 2015. So last year celebrated the 10y year anniversary and at this at almost at the perfect time that we celebrated the 10-year anniversary we calculated
0:54that we've done a billion deployments. A billion is a huge number of deployments especially considering that versel makes everything on that you deploy routable. We host we build we route uh uh you can time travel. If you guys have used the versel platform, you know this benefits. Um, but then since November last year till now, in about 10 months, we added another about 1.4
1:23billion deployments. And even this morning, I woke up really really early and I was asking my agent because the the way that I get this data in itself is really interesting. we have an an internal agent called V that can answer anything about the company and I was thinking about this fireside today or this panel and uh I started asking some questions and uh for for example it was the the
1:48agent was telling me that there's no signs of this acceleration stopping or slowing down if anything it's continuing or that about 90% of traffic to Verscell is originating or basically targeting uh deployment ments made by agents. So basically like the agents have have really taken over and
2:11also have like a life of their own and as you guys have seen right like we just design they can do certain things at this point they can do you know almost every task that that a human engineer could do. Um, and so if this pace of of acceleration continues, I think we're going to see more software being created next year than possibly in all all human history. Now, is it a good
2:37um thing for developers? I I think absolutely, right? Um, you know, one of the other things that I do looking at Verscell and and all of our agentic usage is I try to understand, you know, how much of the work that we do day-to-day has been fully replaced by agents. Meaning, do we have humans uh steering this ship or do we have agents reaching full autonomy? And the answer is we have humans steering these agents. In fact, the word that you guys do see in harnesses
3:04these days is steering. We're always steering these agents. And we're still putting a lot of human effort into making sure that these products are amazing, that the infrastructure is sound and you know, so for as long as that's true, I think it's going to be a panacea for engineers. Also, I caught a little bit of the previous session and we're talking about skills. Clearly, you know, different people get more out of agents than others. And so, it's just like all software
3:31engineering, some of us, you know, produce less or more code than others. Now, you know, with the addition of skills and and how you leverage the harnesses, etc., software, uh, some engineers will get more out of agents and those engineers will be at an advantage. Yeah. So let's let's expand on that. Uh so let's take two engineers. Uh let's say they are equally smart. They use the same models,
3:58they use the same tools, but one of them is much faster and uh better basically. Uh C can you tell me what what that person does differently on a normal day? Yes, I can I think there's quite a few things and I think this is a great question because uh it's almost like the new curriculum, right? If if Astra was to grade me on like how good are you at agentic engineering, we do all have to put our effort into thinking like what are all of the different dimensions that make you
4:28really good at agentic engineering. So one that immediately comes to mind that I've mentioned a few times is what is your ability to parallelize work. I do think that back in the day there was a certain class of like engineering mentality that it was like you go really deep into a problem like I I can almost equate it to when you see the agent reasoning right and the agent is like hyperfocus
4:53and going back and forth. Well, no, the evidence is this and I have this gate and like blah blah blah. Like all of that we used to do as a as our own brain monologue, but we could only do it on one task. You know, no matter what you think about AI, the reality is it can scale way better than us. It can parallelize and multi-thread way better than us. And so I think what makes you really good as an engineer has definitely shifted. I think can you multitask more? Can you context switch?
5:25Can you have a bunch of agent threats? Whether it's investigations or debugging or trying out different hypothesis or or working on different features across git workre whatot there's definitely can you manage multiple agents at once and you know some people have used the cheesy metaphor of like you're now an engineering manager you're not an engineer but I do think there is c something to that right um the other one is how well and I truly believe this is how well do you
5:54understand the model are you getting the most that you can out of this incredible wonder of, you know, it's part database and it's part reasoning engine and it's also someone that can critique your work. And so I I do think there's still a lot of value to the way you prompt it.
6:15So a good example here would be, you know, sometimes you might infuse into the prompt things that are just wrong or you might be too opinionated and perhaps a certain model can actually offer you a bunch of alternatives. And so I think whether you spend the right amount of your time planning with the model and hearing the model out w will I think make a difference between
6:40like the best developers and like the ones that are not getting so much out of the model. And the other one I think that's really exciting right now is okay assuming that models are amazing at like 90% of the task fully autonomous. Are you able and by you I mean your company or you as individual developers are you able to actually leverage that autonomy? A really good example of this if you if
7:05you go and act is Sahil the creator of Gumroad. He's created this little agent that does a ton of his work, responds to every support ticket, and the support tickets that actually are pointing out product defects because like anyone that's shipped a product at scale knows that you have to give support for your product and you get tickets and you can answer those tickets with AI, but some of
7:30those tickets are actually pointing out that your product is broken or that there's a design flaw in your product or that you have to improve the documentation. And so have you been able to deploy what we can call this like a software factory where a signal from the universe comes in and you're able to turn that pretty autonomously into product improvement. I think the engineers that can do that today, you know, Elon calls this the factory is the product because in order to produce
8:00great Tesla cars is not it was not about producing one. It was about scaling up the production line of this of these vehicles, right? And so are you able to create that factory for yourself uh and be really productive and obviously ship high quality products? Yeah. C can you tell me about a moment recently when an agent did something and you thought okay I didn't expect that. And
8:29can can you tell me how that changed how you work? Yes. I'll tell you about a category of things. Uh, so I I I actually tweeted about this a couple days ago. So like I think the intensity and focus in the pursuit of an objective always astonishes me. So a couple days ago the example was I pointed out to this agent that the and this I think a lot of you guys will resonate with this where there's just a long tale of bugs and you're like h I could sol I have the intellectual capacity to solve it.
9:05I just don't have the energy. And so an example here was my website was rendering incorrectly only in inapp browsers inside a mobile screen and just the effort to reproduce and launch the iOS simulator and do all of the measurements, all of the also all of the classes measurements like you
9:33know that the iOS viewport is dynamic. So as you scroll it gets larger and so when you position something you have to be aware and and now there's selectors that make this easier like DVH viewport units whatever but like it's still a nightmare and so that entire class of like software engineering problem which is like we can all kind of intellectually solve it but the energy required
9:59uh is just astonishing and also when you do fix it the product is absolutely higher quality. I love throwing agents at those problems. Like we can call it like sort of like quality assurance and testing and whatever. Models have been getting so much better at vision. Um sometimes I think vision can be slow. So I I'll give you a good example of a good agentic engineer sometimes knows how
10:24to prompt and and create verification loops that are highly efficient. uh and so um it's not just whether a model can do something sometimes but it's how fast can you make the model do something.
10:37And so as an aside uh answering the previous question, the other one so we've been spending a lot of time uh looking into cyber security and spending tokens and money and research into cyber security and the same principle applies when Kim K3 came out um you know we started evaluating it uh against our cyber security benchmarks and what we saw was that Kimmy is just absolutely
11:06relentless in try to exploit uh an objective that you give it. And it kind of scared us to be honest because you know the modern models and harnesses from the big labs in the US kind of hide a lot of their thinking tokens and reasoning tokens and they refuse a lot of uh prompts. So to see the dedication so there's this tale of two cities. You can have a model that's like ultra dedicated in
11:30fixing this trivial rendering bugs, but the same model can be ultra dedicated in hacking you. Now, for every you know sort of bad premonition about AI, there's also upsides like you can deploy those models to protect your code bases. And so your these models will use their immense resourceful in trying to find bugs in your codebase and they're amazing at it. So at out of Brazil,
11:57we have um an internal agent as part of our factory called Hacker. Hacker is doing this but defensively all day and all night trying to find people uh sneaking in back doors into our open source software supply chain or trying to like analyze uh whether the code has any like like edge cases that the engineer or the engineer's agent didn't think about. So everything around
12:26cyber security has impressed me so much. I mean so much so that if you guys haven't tried it yet, I highly encourage you to run deepseec which is our cyber security harness. Um run it on your code bases and it's very likely that it's going to find a lot of stuff. Uh you know there was this meme that went viral in X about like coding is solved bugs are not which is kind of s a silly statement.
12:52Coding is solved bugs are not. There's a truth to this in that if you ask some of these modern models to find bugs, they will find them and then you'll fix them and then you ask again and they will find more and then they'll find more. So pay attention to the cyber security category of bugs because I think it's a it's an open liability for the world and it's something that um you know it's
13:14a good use of your tokens. Yeah. So at Astra we keep cancelling uh SAS subscriptions and we are building our own tools instead. Um have you done the same at Versel? Can can you speak about that?
13:31Uh what did you replace? Yeah, I have a channel on Slack. Sometimes people ask me how I run the company. So I'll give you a little bit of inside baseball. So I have a channel on Slack called product area updates and every week I get a report from every Monday from every product area. This is what allows me to scale myself horizontally by the way like we have a lot of product areas right
13:53we have Eve and we have um uh labs and we have CDN and we have compute. So I get this updates right and yesterday I was looking at our data infrastructure uh team. So this is the team that maintains our data lake. It's maint incidentally maintains the agent that I just told you about the one where I can ask any business intelligence question like what percentage of traffic to the Verscell global CDN goes to deployments created by agents. So that team was giving me an update.
14:27They're cancelling five SAS subscriptions. I I I learned about this yesterday. Five. And what I look for as well is I'm not naive, right? Like I I I actually tell you guys like we all love to create new zero to one things with AI. And there is a downside to that. AI is so good for like that
14:51initial it's a mix of like an IKEA effect. If you've heard about the IKEA effect, people feel better about assembling something themselves than the easier route. So this is something an effect to correct for like it's not always good to create things yourselves. Um and there's also that no software engineering project is just 0ero to one. The it's always the hardest part of engineering
15:21has always been one to 100 keeping it fast keeping it secure keeping it scale. So what I did is like I I started poking holes and see and seeing okay is it right that we're converting these four SAS into things that we've rolled ourselves on top of our cell and the answer actually was overwhelmingly yes. For example, two of these tools were doing the same thing but for different personas. So one of them was like what you would call a no code data transformation pipeline.
15:53And so that kind of software, no code as the name implies, emerged for a time when code was scarce and difficult to produce and difficult to understand. We're past that time, right? Code is basically free now. And so we've deleted a lot of these no code things. By the way, we've also seen that with our customers, right? There were a lot of noode things for building websites and
16:22applications and we've seen tremendous migrations from those solutions to versel. A lot of website builders, a lot of app builders, people are just using coding agents, Nex.js, some framework, whatever for sale. So obviously you're also asking someone that has a lot of skin in the game. I love that this is happening because you know I was literally telling 20 CIOS the other
16:48day Verscell is the last platform you guys need to buy because if you buy Verscell and you govern it and make it secure then you don't need to procure more software and by the way from a security and governance standpoint that's it it it really is freaking awesome because what happens what happened to uh and the reason that the CIO just chief information officer or chief information security officer is that enterp Enterprises were buying and buying and buying and buying software
17:17and they needed to figure out 10, 20, 30 different ways of managing o permissions directory sync and all this stuff and so actually companies are loving the fact that they're strengthening their security posture as they generate more software. Um now you know there's probably classes of software that that you shouldn't generate and um and and what I hear from engineers a lot is like
17:43okay does that mean that like there no one will ever buy software ever again? I really think that if your software is more on the infrastructure side or is providing a platform that works really well with these agents, meaning that you expose an MCP or an API uh and that agents love to use, then your your your software, your infrastructure will continue to be prolific within these companies.
18:11Yeah, we we have 10 more minutes, so I'll I'll try to be quick. So um I want to talk about AI gateway versel AI gateway. Uh you basically see which models companies really use in production. Uh can you tell me what do you see there that people on the outside don't see? Um yeah just tell me about the current landscape. Yeah. Uh, you know, I posted a thing that went viral this weekend about
18:40like 80% of our token traffic is now open models. There were a lot of responses to that that were somewhat like cope or defensive of the AI labs. By the way, the AI labs are doing great. Like all of AI is doing great. Astra, open AI, Astra, uh, Anthropic, Verscell, like everyone that's in the token economy, I think, is doing well. Like I think we can't, no one can really complain. Um,
19:07if you're creating value and you're disseminating intelligence and you're making things that before took lots of teams of people and like now you're making them easier, you're probably doing great.
19:18But I'll say that people were saying, well, yeah, because open models use more token or open models do this or do that or but the spend. Well, no. The reality guys is when we started the AI gateway, it was all open AI and then a lot of it was Gemini and then a lot of it was anthropic and those three have consistently been losing ground to open models. And so that doesn't mean that I'm
19:45bearish on them by the way. Like I would I would actually like if I if I if I were to buy stocks and this is not financial advice. It's like anyone that's in the AI world at the different layers like I would have them in a basket of like AI goods. I love AI. AI, you know, um is is again is doing amazing. So, but I'll tell you open models and faster models seem to be the new
20:10trend. The excitement about Jev points out that people need high quality, fast decision making and they weren't getting that from the big models. And so, a lot of people are saying, "Okay, get it.
20:28I'm using AI but I'm optimizing. Now another thing that AI gateway is offering companies is just the feedback loop of like do you know how much of your how much your engineers are spending every day like how much you're personally spending uh do you have budgets in place when you issue an API key are you putting a budget on it um and so AI gateway is doing this like costgovernance task
20:54that I think is really important because it's really easy to just set your coding hardness is to fable 5.1 extra high fast and forget about it. And so what the data is showing us is that people are moving past that sort of token maxing use AI at all cost mentality and they're starting to optimize into you know combinations of models. Um the other thing that the data shows us is that
21:22when a team adopts AI they always end up adopting multiple models. I think the average enterprise is like over five. Um and uh enterprises are now getting really savvy about evals. Uh this is also inside baseball about versel is we're almost like an eval first company now. When we develop nextjs we start with evals which is kind of nuts like used to start developing a framework by I don't
21:51know like writing tests writing software maybe writing documentation like different approach. No, it's all eval first now because we care about how the different models and agents use our products, right? When we design APIs, we're eval first. And so I think um the world has moved into this eval first like sort of or is moving towards that and that's going to help make more intelligent
22:14decisions about models and um and right now you kind of have to navigate this fog of war of um like vibes. A lot of a lot of us are still deciding on models based on vibes. Oh, I saw on Twitter that like Astra is better. It's like is it like some of our Astra came out and they moved back to soul for certain things. So I think the TLDDR is that we're seeing tremendous model
22:40diversity. Yeah. So now uh imagine that you are 20 like let's imagine uh you live in Ljubljana the capital of Slovenia. you have a laptop and it's September 26. What do you start doing or building?
23:04So, I still think that domain expertise in in any area is extremely valuable, right? So, let's say that you're a person that says, I'm going to I'm going to get really freaking good at developing coding harnesses. Great. Like, go really deep into that topic. And now you have immense leverage from the ability to automate your software development. I think this is true in every area, right? Like I think you guys are such a great example of like education. My god,
23:33like there's probably going to be like trillions of dollars of value to be created still in the area of like educating ourselves better, our kids better, our families better with AI and so on. I think the space of personal agents that people can really trust. Trust is an area I think that needs a lot more investment. We're now seeing sort of like the big company agents come out,
23:59but I'm a I really believe in open and I really believe in data ownership. Something that like um you know, Rouch G Europe, Handshake, we agree on is I think the value of data sovereignty. And so I think um yes there's a lot of like proofs out there in the world that AI can do amazing things but maybe it's not in the way shape or form that customers and consumers and enterprises will want
24:24it in the fullness of time. In fact AI gateway is a great example of this guys like obviously like the world wanted AI but it came with all of this like tradeoffs when you were actually going to use it yourself. Like you know there was a big model outage yesterday. So AI right now is in this like tale of two cities of like I see miracles every day and yet I don't see enough people using them
24:49using AI uh especially in like worldwide right or I don't see them using it effectively enough or cheaply enough or locally enough. So I think there's just you know sometimes it can be I can feel that you know some someone says oh my god the agents can do all of these things engineering is over and and I actually see the opposite like 95% of my day I see oh my god like we haven't put it
25:15in enough places or it's not fast enough or it's not tailored enough to each user um and then like I said the domain expertise like uh you guys have seen some of the greatest companies out of Europe have come out recently in my opinion uh out of focusing on a on a vertical whether it's legora and law Astra and education you could argue 11 labs and voice just so going deep in in one topic
25:41is really helpful also as a as a person you can actually now straddle your skill set a lot more I I just made a small investment in an European company where I saw that the founder was just really cracked at how he was like representing his brand in socials, creating videos like cool ways to present his company. The product looked really cool. He was innovative in an area that
26:07I think is really cool to innovate with AI in which is go to market. Interestingly enough, from my vantage point, San Francisco Silicon Valley, a lot of us are still doing go to market a little bit like it's 1986. Uh contact sales and like all of this stuff is like it hasn't changed as much as you would think. So if you if you raise even five inches above the bubble, you will see
26:30that a lot of categories of software and areas of professional work and life are still relatively underdisrupted relative to the intelligence that we now have on tap. Certainly. Awesome.
26:45So uh I I have the last one, the last question, Jarmmo. Yeah, go for it. So one year from now, it's September 27. Uh let's say a team of five people, what does their week look like or what they can can build? They they they cannot build today. My sense is that those five people will likely be working on sort of the brain of the company. You you can call it factory,
27:12you can call it the super agent that runs the company rather than any of the individual tasks themselves. So maybe to give you another example from the day-to-day of our cell, our design team used to work on design artifacts. Not the case at all anymore. They work on the design tooling that yields those artifacts. So one that I had shown publicly was anytime we do anything for, you know,
27:39we we just uh signed a a major partnership with Mercedes F1. So, if you're a Kimi fan, smart, congratulations. We're winning all the time. Uh, so Mercedes and Verscell are are going really deep. That means that we have to do a lot of like marketing stuff like there's a new race and we have to create a poster. We organize a dinner with people and we send out invites. our ability
28:04to reach all those places and and and do all of those events and show up with a beautiful brand to all of the places that we go used to be limited by how much design headcounts we had. So if I sign a a partnership with Mercedes, okay, I have to hire another seven designers because we used to live in this linear world. We're going into a nonlinear world. I could imagine those five people
28:30producing so much output that you would be like, "Oh my god, that used to take an entire publicly traded 1500 person team." That's why I think small teams, if you work in a small team, you're you're at a surreal uh asymmetric advantage in the new world. Uh and so maybe to finish that story. So, I just saw the latest iteration of this design factory that my team built where they managed
28:57to put themselves in the loop of this agent to do the final sign off. So, they're giving an agent to the marketer to myself if I want to. I'm about actually I'm literally about to tweet out about a a fireside chat that I'm going to do with Toby from Shopify. and uh I have to get a sign off from the design team so that I don't ship slop to the world. And so this agent is automating the entire
29:22process of ideiation, rendering, feedback and forth pulling templates. Think of it like a design harness and the design team is in the loop to do the final sign off so that we don't ship slop. So it's like the entire universe is in balanced. Uh Jermo, um thank you. This was awesome. Uh, next time, uh, let's do it in person in Ljubljana. You can come to our unicorn party next year. Okay.
29:51Thank you guys. I appreciate your partnership. Uh, I'll definitely be there soon. Uh, one fun fact, I'm a huge fan of Alone. It's a survival TV show and uh uh uh the I think one of the recent winners was from Slovenia. That guy was so cracked. Uh so keep keep doing uh great work and uh stay connected online. So send us feedback on X and engage with us. Uh we're happy to um to keep this connection going. Thank you so much. Take care. Bye-bye. See you.
