Viktor Revenue, Valuation & Funding (2026)

Nathan Latka (00:01) Hey folks, my guest today is Fryd Wiatrowski. He's an Oxford trained mathematician and computer scientist who previously worked at meta AI and in high frequency trading before co-founding Zeta Labs in twenty twenty three. He's now building Victor, an AI coworker that lives inside Slack and Microsoft Teams and has become one of Europe's fastest growing AI startups. Frederick Roda takes us the top. Fryd Wiatrowski (00:22) Yeah, nice to meet you, man. Super excited. Nathan Latka (00:25) Good to meet you too. I have to ask you first, you're hearing these stories about like Zuckerberg paying, you one engineer a hundred million bucks. The talent wars are real. Why did you leave Meta AI? ⁓ Fryd Wiatrowski (00:34) so I was not in Meta AI specifically, I was in Meta. and I was just so I first interned at Facebook, that was in twenty twenty two. I was I was w working with distributed logs and then I joined full time, ⁓ got this full time offer, joined full time, stayed for six months and left. you know, I I I I loved Meta. It's but it's not for me. Nathan Latka (00:58) Fair enough. Okay. I w I wanna I wanna talk about what Victor is today, then go capture your backstory because it was quite a journey to get here. So tell us what you're selling today. What are the big use cases here? Fryd Wiatrowski (01:07) Yeah, so Victor is your AI employee. It's ⁓ it's it it it it's AI employee to an extent that you know it's really difficult difficult to distinguish Victor from like a horizontal hire that you just make. You commu you communicate with Victor in the exact same way as you communicate with your teammates. That means for Slack. You don't need to go to a separate web app and switch context to speak to Victor. Victor works across various verticals, you know, it's ⁓ just like most language models today, it's knowledge is horizontal. It's parti the the three biggest verticals that we have are e-commerce agencies, particularly marketing agencies and tech startups. in terms of the use cases, you know, a anything across data entry, research, ⁓ or or even coding and building building apps, which is by the way very powerful for Victor, ⁓ yeah, those are the biggest. Nathan Latka (01:57) Mm-hmm. And how are you pricing for this? You know, I had Amanda on at One Mind. You she's the previous founder at Six Cents, you multi billion dollar company. She's now building One Mind. It was just kinda it's not it's not really in your space. I'd say you're very different, but she's actually billing these. She's charging groups like HubSpot like an employee. They're paying a hundred grand a year to like use this virtual employee. How are you pricing? Fryd Wiatrowski (02:16) So we have ⁓ subscription tiers and they're almost uncupped. You can pay as you can see, ⁓ up to a hundred thousand dollars ⁓ a a month. And the the way it works is you start with free credits. You basically ca I I think there's like a hundred dollars in free credits currently. ⁓ and the moment you run out in Slack, Victor will tell you, Hey, I run out run out of fuel, can you upgrade? Then you upgrade to next year the next year, and then you Yeah, exactly. ⁓ Nathan Latka (02:42) Like here. Wait, this is crazy. Do you do you have people paying you fifty thousand a month already? Fryd Wiatrowski (02:46) And then Yeah, we have ⁓ we we have like five five person team paying twenty K a month. Yeah. Yeah. no no, that's not the largest customer. It's it's one of the, you know, one where the the size is size of the subscription to the size of the team is the largest, I think. ⁓ the you know, it it is, it is. I remember when Chat GPT Pro launched with for like two hundred dollars, it felt so huge to us. Nathan Latka (02:50) What? What okay, is that your largest customer? That's insane. Fryd Wiatrowski (03:16) That we decided to have a shirt subscription for the whole team for like you know, ten people back then. And now now we have this tiny team paying twenty K a month for Victor. Nathan Latka (03:28) That's why. Okay. So you're pure there's no sort of gimmicks here. It's really like if you use more credits, you pay more. Yeah. Okay. And some Fryd Wiatrowski (03:33) Correct, yeah. And then if you don't want to upgrade to a higher tier, you can also top up one of. Nathan Latka (03:39) Mm-hmm. Mm-hmm. One of the one of the knocks or that people are trying to figure out is like, man, you know, are the found is the foundation model going to replace this sexy new AI tool, right? And one way to look at that is like, you know, how much of your cogs are just going to the foundation model? Can you talk on that a little bit? Fryd Wiatrowski (03:53) Hundred percent. Yeah. So so typically we try to have ⁓ like a thirty three percent margin. So we take what we pay for entropic or whatever is underlying, we multi multiply it but multiply this by one point five and this is what we charge. ⁓ so we try to maintain this margin. ⁓ this is what our credits are priced. Nathan Latka (04:09) This makes a ton of sense. Okay. So that gives context on pricing and sort of where you're at today and how many customers do you have today? Fryd Wiatrowski (04:16) twenty eight hundred paid customers or twenty nine. Yeah. Not many. And and the revenue is high, right? so so Nathan Latka (04:20) And that's wild because the Yeah, well you you gave a good we're we're recording this on June second. You gave a great article to Fortune where you said that you'd broken a fifteen million run rate in two thousand organizations. So but that was like forty-five days ago. Are you comfortable sort of sharing what you've done over the past forty-five days? Fryd Wiatrowski (04:36) Yeah. Yeah, so so in the in the last month we added ⁓ two million in in in run rate or like two point five, something like this. ⁓ so we added a lot of a lot of teams. I think like we added so back then when we shared with Fortune, I think we had like two thousand, now we have twenty nine hundred. ⁓ but there is a lot of one person teams. So basically people creating dark workspaces to use Victor. And therefore the RPU on those is much lower. Nathan Latka (05:02) Yeah, but you have a nice range. I mean, you could have people at twenty bucks a month and you have what's your I are you comfortable sharing? Don't share the logo, but what's your largest customer pay? Fryd Wiatrowski (05:10) so as of today I think it's like thirty to forty K a month or something. Yeah. But those are those are small teams. Those are small teams, right? Nathan Latka (05:13) Yeah, but that's a huge range. Yeah, yeah, yeah. Interesting. Okay, let's get I want to get your backstory here. Okay, well, I just dove rope into the numbers because I was curious, but you didn't just obviously come up with us and do your thing. Tell us about this. You you leave Meta AI. What do you launch in twenty twenty three? Or even before that? Fryd Wiatrowski (05:29) So back then, twenty twenty-three, you know, when language models like you know w when ChatGPT launched went super viral, it was clear that ⁓ it's not gonna stop at at answering questions. And back then you didn't have like reliable code gen or tool calling. so the only way for models or like AI to take action in the real world was for the browser. So you basically take a snapshot of the DOM of your HTML. you compress it in a lossless way somehow to make it fit into the 4K context window. And then you give it to the model and decide and and then ask, you know, this is my objective. This is this current state. What is the next step? And you do it in a loop. and that was like kind of the best the first way to build agents. You c you probably remember the the auto GPT moment. I think it generated a lot of virality because of the promise, but it just didn't work reliably. And this is also what we were struggling with early on. Yeah, those agents. Nathan Latka (06:27) So it's really the it's the context windows were so small. You had to sort of g gimm be gimmicky, sort of bolt it together a little bit. Fryd Wiatrowski (06:32) Yeah, so so we had to compress the DOM because HTML was very extensive. ⁓ so we had to compress it to to kind of maintain what only what's what's relevant. the problem is even if later on we had like thirty thirty-two K context windows and a bit larger, but that didn't work because with the size of the of the context the the the reliability was dropping. ⁓ and and so over the whole twenty-three, twenty-four, you couldn't really build reliable browser agents. But then in twenty-four If I remember correctly, ⁓ you know, Sonnet 3.5 launched. And then we were able to build the first agent loop ⁓ that was able to make tool calling. And our and and this is when we launched ⁓ so we started building. We didn't launch. We launched JS AI, which was like an email agent. We launched it in February twenty-five. ⁓ and basically what it what it was, ⁓ it was ⁓ so we gave up on the browser automation thing because we thought this is not not reliable. You know, if it does four steps with the reliability of sixty percent. It doesn't make any sense for anyone to use it. It's a fantastic research research project, but it's just useless. So we couldn't we we would probably have died if we continued doing doing this but at that moment. So we Nathan Latka (07:42) And that was the twenty twenty three to twenty twenty four project before Sonnet three five. Fryd Wiatrowski (07:47) Yeah, then Sonic Free5 comes, tool calling, you know, agent loops ⁓ possible now. So suddenly you can perform some more complex actions. and so then what we did is we basically placed it placed it in emails because we really didn't want to build the the kind of an interface where you need to go to this agent to ask it to do things because we observed that the creativity of users back then was quite limited, of like what's possible. And and and because the agents were not not a thing. So we we thought, okay. It's probably best if the agent suggests the automations themselves and people just approve them. And where where can the agent find the most automations? Well, in emails, right? So we place this agent loop in emails. Basically the way it worked is first the agent was labeling the email, whether that's a promotion or like an FYI or it needs a response. And in case it needs a response or it requires some actions, we triggered the agent loop. And the agent loop has an objective which is determined by the what's inside of the email. Like for example, hey, give me a refund. The agent loop's objective is then go to Stripe and do a refund. And the Stripe must be already connected by the user. ⁓ and then also we were indexing the emails of the customers. Yes. we were indexing the emails of the customers, and then for every email, whenever whenever an email arrives, the agent loop is triggered, and it typically outputs the draft as well. Nathan Latka (08:57) Did that work, by the way? Fryd Wiatrowski (09:10) we first gather the context from the ⁓ from the SpectrumDB by proximity and and then we based on those emails from the context we craft the the the the response which for example if someone asks you about the details from the invoice which you had somewhere received somewhere last year, Victor or like Jace back then, Jace will go to this particular email in the history, open this invoice and give you those details in the draft. So in certain scenarios it saves like hours. Or for example, when it you need to collect the invoices from the last month. Someone emails you, Hey, can you send me all the invoices, like your accountant? And then you open your email inbox and then suddenly there is a draft with all the res all the kind of ⁓ invoices attached, which is kind of crazy. It it felt. It felt. ⁓ the problem here for us and why we couldn't scale it really was the costs. Because we didn't have any margin at all. Because we while for for for typical agents you need to trigger them. This agent yeah was triggered automatically. yeah, so basically everyone receives emails all the time. So we constantly trigger this agent loop. And because it's not just a prompt that gives you a draft, it's actually, you know, this agent loop which is very expensive, ⁓ we couldn't find the positive positive margin there. ⁓ I think we didn't try hard enough with usage based pricing or like what we do with Victor right now. But that's, you know Nathan Latka (10:11) It was constantly running. Fryd Wiatrowski (10:35) That's that's too late. Nathan Latka (10:37) Well, let me so let me ask you a follow up question on that. ⁓ you know, we just acquired a company at Founder Path and we figured, you know, and we're not getting any of the team. So what we did is we went into Google Vault, we downloaded like literally six hundred and eighty-five thousand emails in an Mbox format to try to create a brain of the company we just acquired. ⁓ now you know all this, you're an expert. My audience, maybe if you're not following along, we're just we're trying to get the context of how these people emailed customers, right? And so, Fred, what I'm hearing you say is basically You wanted to give people the ability to understand all your historical emails, to write the next email you needed to send, but every time you ran that loop, you just couldn't make the margins work and that's why you stopped it. Is that what I'm hearing? Yeah. Inter So why not just charge more? Fryd Wiatrowski (11:19) Yeah, probably that it it felt ridiculous to for us to charge more than sixty five dollars. But then we were charging sixty five on like the pro plan. And we then we had like and people were like raging, like this is so expensive, what the hell? And you know and and like, you know, we we tested higher prices, but higher prices for us meant sixty five. ⁓ so and in the in the end like I think it's to be honest, I think it's just my mistake. Like should have pushed four usage based and and probably we would have Nathan Latka (11:27) my God, I would pay a lot. Really? Interesting. Fryd Wiatrowski (11:49) found something now it's usage based and we have left it, we don't do any marketing, we still maintain the product, ⁓ and the revenue is not dropping. So it turns out that people are willing to pay more. Nathan Latka (11:59) Mm-hmm. Interesting. Well, look, the reason I wanted you to tell that story, there's a lot of people listening right now building their own agents and wrappers, and they're not quite sure how to make the margins work. And guys, if you don't make the margins work, you're just literally digging your own grave because you're not making any money. Who cares how cool it is and how much you're on product on and TechCrunch and Wall Street Journal? Like if not making money, you can't build something sustainable unless the credits on the models drop. So, like Fred, I I have to ask you these questions because you're a technologist at heart. Inference models, right? Tool calling into open source models on Hugging Face. Like, can those credits be more powerful than using a sonnet four six or four seven credit for certain tasks, then are you doing that to optimize your margins? Fryd Wiatrowski (12:32) Yeah, so so the last time I digged into this, you know, ⁓ the the the the the the most optimal solution for us was using Opus or Sonnet, because they have great caching. and with the op with the open source models back then, you know, it was just we we could have like fifty percent caching and but not it was not that that great. and similarly, you know, currently we are testing open source models for Victor. So you can go to Victor and Nathan Latka (12:43) Mm. Fryd Wiatrowski (12:59) you know, ⁓ or and use an open source model like like Kimi, but they are just not performing as well as ⁓ as Opus or GPT five five point five. Nathan Latka (13:09) Interesting. Okay, but it's fair to say if I looked at your PL from last month, I would see the majority of your cogs for the foundation models would going towards anthropic. Yeah, yeah, interesting. Okay. ⁓ how are you deciding what jobs to be done you should keep training Victor on? Like you it sounds you start off with sort of a marketing and sales approach, but as you said, this could be used for anything. Fryd Wiatrowski (13:17) Yes. Yeah, the the first kind of the first story of Victor is when we we first built it, we added it to Slack and we we we just had internal issues with ⁓ with marketing and our spend and you know some the CPA has gone up like crazy in meta ads and and and we just added Victor as a test connected to the meta ads to to our meta ads and then Victor randomly says, Hey guys, you have the the audience network toggled on You should probably toggle this off. And then we did, and we started saving like ten K a week on like it turns out that we were spending we were spending ten K a week on audience network. And you know, I had great people auditing our other account. And the reason that they didn't notice that is that this setting is buried very deeply in the UI. And that's, you know, very difficult to find for a human, but for an agent it's just a list of endpoints, right? And the the agent sees, shit, there's something Nathan Latka (14:10) Mm-hmm. Fryd Wiatrowski (14:28) ⁓ something something wrong here. And so Victor suggested this and we immediately saved like ten K a week on marketing spend. ⁓ because we didn't throw it away for audience networks. We audience network means advertising outside of meta through meta ads, which makes no sense for us. ⁓ so yeah, that that was like the first internal PMF. Like suddenly everyone started speaking to Victor and using it for for all their use cases, whether that's in finance, operations, marketing, product, essentially everything. And then we just wanted to ASAP. Nathan Latka (14:58) give you a prompt and you tell me how close you think Victor is to be able to able to solve it based off the example you just gave, right? I'm a software founder listening right now. Hey Victor, I want you to spend $100,000 per month on paid Google ads. I want you to op you know my ARPU, you know my average price point. I want you to optimize for a 1.3 X ROAS where I get 50% of my CAC recovered on the initial checkout. And then I want a two month payback period optimized. Run a thousand tests until that you can make that closed loop work. And then once you make that closed loop work, start driving my RAWAS up by 1% every 30 days. So a fully, truly closed loop system, how close are we to that? Fryd Wiatrowski (15:36) ⁓ I think that Victor would probably start doing branded ads here. Which is kind of ch i i it like it would Victor would inform you, hey, you can easily achieve that with brand if you're not running brand ads. So that's the answer and then you know it can easily do it. ⁓ because brand is cheaper. but then if you tell it to spend it on non brand as well, ⁓ yeah, I think I I think that's that's achievable. On the numbers, you know, that's also very ⁓ it's a derivative of your product. Nathan Latka (15:45) Yeah. Mm-hmm. Fryd Wiatrowski (16:09) So Victor can't fix your product issues. So if if your product is not great, if you don't have a PMF, then you know, Victor will run some experiments but it will struggle. In terms of the execution itself, yeah, that's super easy. ⁓ you just schedule ⁓ you can you can schedule ⁓ like daily checks, like a Chrome with daily checks and then rerun the campaigns, ⁓ kill the the losers, ⁓ scale the winners and and that's it. Nathan Latka (16:36) If everyone's running that same optimization process though, there's only a set amount of ad inventory. Doesn't there isn't it? It's not it's a zero sum game. If somebody's winning, somebody has to be losing. But if they're all using Victor, how does that work? Fryd Wiatrowski (16:51) So I think with marketing specifically, on ⁓ on Google ads, when you're bidding on keywords, I I think i if everyone is using the same agent, ⁓ you're probably yeah, we're probably someone needs to sacrifice a lot of money and build somehow an economy economy of scale where everyone else is losing money and then they can only win. on on meta ads where creativ creativity matters, like it's impossible to to have like like this is where start. cre creators are the are the most important, right? And it's in it's impossible that all every everyone will run the same creatives. and similarly not everyone will have the same funnel in the product. So Victor will probably start optimizing those funnels, but the creatives will always differ and it's all about the value proposition. And I think it's still marketing can only be good to some extent. It's just the just about how good your product is in the end. Nathan Latka (17:47) Yep, interesting. Okay, let's talk just more about the growth rate and sort of how you have fund of the business. Obviously credits are not cheap to these foundation models. So when did you do your first round of funding and how much was it for? Fryd Wiatrowski (17:57) It was 2023. We raised what the first check was Nat Friedman and Daniel Gross, and that was 1.5 mil, and then we raised a bit more, like 1.4 from ⁓ our European investors. ⁓ that was PECVC or like Early Bird Digital East back then, and and Kaya VC. then 24 we raised ⁓ another 1.5 from Leone's Capital, and then ⁓ Last year we raised there was a seed round again from the same investors Beg VC, Ca VC and Innova VC. And so so today to d until the series A we raised like what thirteen million dollars. Nathan Latka (18:41) Yep. Yep. And that w the the seed round, sorry, was a four point five million seed round last year. Fryd Wiatrowski (18:45) Four point five. No, sorry, the seed was eight last year and twenty five. Previously four point five raised. Yeah. Okay four point. Nathan Latka (18:51) I see. Okay. And then take us into the Excel story. It sounds like they loved your vision. They said we gotta fly out and meet this guy and get this deal done. ⁓ it was a $75 million deal. Look, I'm curious how transparent you're able to be here because this is a really good lesson. And I rarely see this from people not in San Francisco. You do this from Warsaw. Driving FOMO like this with investors to the point where they'll fly halfway around the world to get the deal done is is rare, right? So what did you intentionally do to drive that FOMO and get this thing closed? Fryd Wiatrowski (19:20) So we didn't so first we didn't have any urgency to l raise back then because Victoria was blowing up. We saw that it probably probably better to raise later. back then we were at like what one to two million in ARR. ⁓ Nathan Latka (19:32) What year? Or month? What month was that? Fryd Wiatrowski (19:36) This was this year in like what March? yeah, so it was like I don't remember exactly, but like it was in March. ⁓ end of March maybe, yeah. like one to two million in ARR. ⁓ and and yeah, we we we saw that we are you know, we will probably hit like a hundred this year. So a and we didn't need ⁓ a lot of money for marketing, so we didn't want to raise. But then I had one call with with Jenya, it was a great call. ⁓ Nathan Latka (19:39) March twenty sixth. Mm-hmm. Fryd Wiatrowski (20:06) They really wanted to come to Warsaw and to visit us in the office. So, you know, ⁓ you know, we agreed, let's just catch up. We had lunch together, was awesome. Then then we had just one call and they gave us a term sheet. And, you know, I think they're just great investors. I think, you know, probably are dream investors in Europe. ⁓ so I was very excited for this. Nathan Latka (20:30) The deal was a seventy five million dollar ⁓ series A. Are you comfortable sharing what the valuation was? Fryd Wiatrowski (20:36) Yeah the Yeah, the valuation is is four fifty. Nathan Latka (20:39) Okay. Does that feel high or low to you? I mean, did it in in the moment did you know, what were you feeling? Fryd Wiatrowski (20:46) So when they first asked, so we we originally didn't want to raise and I was very transparent about this. I thought that, you know, probably fun to talk, but we probably don't want to raise now. And then we discussed what valuation would make sense in case we assume you are already at the numbers when you want to raise. And I randomly said, Yeah, that's just like four hundred or something, but that's impossible. but then they are ⁓ like an ha h an hour later. Nathan Latka (21:14) You you said let's do four hundred and they left the room. Fryd Wiatrowski (21:16) Yeah, I I I I was assuming this is like a ridiculous number up for like a two million ARR, like one million ARR startup. ⁓ and so I was just, you know, we just left the room. ⁓ and they're calling us an hour later and they're said, Hey, could you come? And there's there was like a term sheet. ⁓ yeah, I think they they they played it really well. Nathan Latka (21:34) Your growth rate though backs up like sort of this crazy valuation, just to confirm, you know, two million of revenue March twenty twenty six. By May, a couple of months later, you're fifteen million of ARR. And now today you're telling me the past thirty days you've added two point five million of new ARR, right? Fryd Wiatrowski (21:48) Yeah. To be precise, annual run rate, which is thirty day annualized w in terms of ARR, ⁓ it's like what, thirteen point eight and then the rest is top ups because top ups don't annualize. So you d take the last thirty days of top ups and you can annualize them and then you're at like seventeen right now. Nathan Latka (22:07) Yep, yeah, yeah. Yeah. So obviously the growth rate is insane. What is driving most of the growth here? Is it just pure product led growth or is there something intentional you're doing here? Fryd Wiatrowski (22:15) so primarily most of our growth is upgrades. ⁓ it's not the the first subscription because the first subscriptions they are at like fifty dollars per month and our ARPU currently is at four hundred. So people need to continuously upgrade. And therefore the the the the revenue retention is amazing. We have like after the first month we have like three three hundred and fifty percent. ⁓ and that's stabilized. Nathan Latka (22:38) That's ins that's insane. Three hundred and fifty percent net dollar retention over a thirty day period. Fryd Wiatrowski (22:44) ⁓ yes, so so that I like yeah, I I I think that's that's the average right now. I might be wrong. Yeah. Nathan Latka (22:50) It it's like on average if I sign up for a dollar today, by the end of thirty days I'm paying you three dollars and fifty cents. Fryd Wiatrowski (22:55) Yes, yes, yes, yes. so so so then, you know, the the RPU stabilizes at four hundred currently and and what's surprising is I was always thinking that Victor kind of spreads virality in their organizations, which makes sense because you add it to Slack, you use it in public channels, people see that you use Victor, they want to use Victor as well, they do. but turns out that our penetration of the workspaces is not huge. We have like a ten percent penetration currently, unlike Nathan Latka (23:22) How do you know that? What does even a workspace even mean in today's world? How do you know you only have 10% of it something? How do define that? Fryd Wiatrowski (23:27) So so basically look at the number of people in the Slack workspace and then look at how many interact with Victor by either mentioning Victor or DMing Victor. Nathan Latka (23:30) Okay. I see. Interesting. So 10% hundred percent company, only 10% are engaging. So there's more wallet share there. Yep, interesting. And how tell me your team. How many people are full-time today? Fryd Wiatrowski (23:47) ⁓ so currently we're at like fifteen full time where we have like five engineering or like six engineering, one designer, ⁓ three people in customer support, two full time growth people. Nathan Latka (23:50) Fifteen. Guys, if you're loving this story, we want to give Fred some love. He's being very vulnerable. So he's trying to hire. Right. If you're if you're a Fred, give the pitch here. If people want to work with you, what should they do? Fryd Wiatrowski (24:09) I love people who act. So if you can send me outcomes and show me what you do and do something already without even being in the company, that's best. Nathan Latka (24:10) Yeah. What what role are you just like s you're you're like dying to you absolutely need to find this person? It's a huge opportunity. Fryd Wiatrowski (24:24) I think always engineers and and and product managers. I think that these two roles are converging. ⁓ so it's like one role essentially. ⁓ and then on the on the marketing side, on the marketing side is huge as well because we need, you know, performance marketer marketers, influencer managers. ⁓ ahead of conversion is huge because currently, you know, we have quite a remarkable spend on on meta ads, but we don't even run many A B tests on our landing page or in the funnel. And and these two are not great. When you look at our funnel there is just so many mistake mistakes to be fixed. So ⁓ so I think ⁓ it it's very so it's hard to say after two months. ⁓ three months. currently we are spending like what, thirty K a day, but that's in like an experimentation phase. Nathan Latka (24:57) What are you spending monthly on paid ads? Yeah, yeah. Well, guys, there you it. If you're if you're an absolute killer, go analyze all his ads. Write him a ridiculously intelligent report on what you would optimize. And there, you know, who knows? Maybe he says yes. Fryd Wiatrowski (25:21) Hundred percent. Yeah. And the people don't do it. So if you do it, you stand out. Nathan Latka (25:22) All right. Very a hundred percent. All right. Tell us more as we wrap up here. Where what's the product vision here? Where are you taking this thing? Fryd Wiatrowski (25:30) I think it's very clear for everyone that in the next five years the knowledge work that we know is gonna go for a big revolution. ⁓ and I think the size of that revolution by in in impact is probably much ha much larger, not only due to the population size ⁓ of humans, but like in general it's much larger than the industrial revolution. so there is a company or like a set of companies responsible for that revolution and we want to drive this. we currently see a lot of push towards personal agents and like personal tools with AI. We haven't yet seen agents that are team native and spread ⁓ that that are shared internally that have contacts for just like any just like an employee in your company has contacts from various tools and can be contact contacted by anyone in your company. You know, with most agents you everyone needs to set up their integrations, everyone needs to set up their own off. ⁓ it's not you you currently there is no agent where you c it lives in the workspace, everyone can speak to it and can share knowledge with everyone else. It has a lot of independence. My belief is that because of the shared access in the organization, those agents, the workspace agents, they spread much faster than personal agents. And there is no friction in interacting with them. It takes one person in the team to create this agent. Anyone else can immediately speak to this agent. And so the the kind of it spreads much faster than those personal agents. And therefore my hypothesis is that the the the the the revolution, the knowledge knowledge work revolution is gonna be caused by such agents rather than personal agents. Nathan Latka (27:02) Do you have any idea if I asked how many jobs do you think you've replaced today, could you give me a number? Fryd Wiatrowski (27:07) In terms of like how many people lost their jobs because of it? Nathan Latka (27:09) Yeah. Well not necessarily lost, just like you told me about a real estate developer that you helped save four million bucks, right? It's like did you ⁓ you know, it him using you, is that equivalent to two full time people he'd have to hire without you? Fryd Wiatrowski (27:19) So I hear like, you it depends on the company. Like we had one guy who just said, Hey, I didn't have to hire ten people now because Victor did took all those jobs. Or or you know,...