Lyzr AI Revenue, Valuation & Funding (2026)

Company ID

50364

Website

lyzr.ai

Transcript ID

3632 Nathan Latka (00:01) Hey folks, my guest today is Siva Surendira. He's the CEO of Lyzr.ai, a two-time founder, previously built Power Up Cloud, an AWS consulting firm, which was acquired in 2019, where he then ran the AWS business group globally. Today, Lizer builds enterprise infrastructure for deploying and governing AI agents, pursuing a vision it calls organizational general intelligence. Siva, you ready to take us to the top? All right. Siva (00:26) Absolutely. Nice to meet you. Nathan Latka (00:28) It's good to have you on, and we were talking before the show. You know what to expect. It sounds like you you caught some early episodes back in the day, huh? Siva (00:34) Absolutely. I mean I've been like I mentioned earlier, I'm a big fan and a lot of during the early stages of building Lyzer, a lot of the videos that you did was extremely helpful. kind of reaffirmed the confidence that a founder needs, to be honest. so yeah, it was extremely helpful. Nathan Latka (00:52) Well, take me to your backstory before we jump into Lizer. So I'll I'll show your LinkedIn here in a second, but where did you come up with the idea for this? Siva (00:59) Yeah. so after I sold my previous startup to this large global system integrator, I I sold Power Up Cloud. Power Up Cloud. This is the company that I built and ran, and it was an all-cash deal, which is good for me and the whole team. And we sold it and I moved to the US as part of the sale in 2020. And I was leading this global AWS business unit at LTI. Nathan Latka (01:05) Which which one? power up cloud, okay. Siva (01:28) Which was doing sixty-five million when it took over the whole business unit. And I did about I took it to about two hundred and twenty million in under two and a half years. So did about four hundred million for the business unit in three years. And during that time, I used to go into this QBRs with Excel sheets, spreadsheets, and pivot tables, and only for the leaders to ask a lot of questions, which I'm not prepared at all. I mean, I know my business really well, but The line of questioning was very different from the management team. And so when my time was up at LTI, which is when my golden handcuffs came off, I wanted to come back to building startups. and the timing was perfect because that's when we had the GPT 3.5 API come out. So I obviously, as a big data engineer, I started got back to coding, started building few systems, and I realized, okay, what if we address The problem that I faced, like Paul Graham says, right? If you're trying to find a problem to fix, try and go back in your own life and see what was the problem that you are facing. So I thought, okay, what if we build an AI data analyst that can not only analyze your spreadsheet, but also can prepare you with all these questions and ideas that you may have so that you can understand the data really well. That was a whole thought process. And when I spoke to a lot of these vice presidents in large services firms and banks, I realized that all of them, all these folks are on the same boat. None of them have business analysts or dedicated EAs who does the work for them, which is exactly where AI can play a massive role. So that's the whole background of where the idea for Lyzer came from. So Lyzer in its infancy was an AI data analyst platform, right? While building it, I realized that there is a bigger problem to solve because even as a solo developer, I found it very hard to deploy and manage all these open source packages. And for enterprise, it would definitely be a nightmare. So that insight led to pivot in the first three months of starting Lyzer to becoming an enterprise framework and platform, allowing enterprises to build AI systems and agents. So that was the backstory of where Lyzer came. Nathan Latka (03:52) Let's take a step back for a second. What year did you write the first line of code for the platform? Siva (03:57) which year you mean? Twenty it was April twenty twenty three. Nathan Latka (03:58) What year? Yeah, twenty twenty two, twenty twenty one. the first line of code was April twenty twenty-three. okay. And you you have a ton I mean, when founders come to, you know, Founder Path to raise from our fund, I go, okay, if it's a solo founder, great. They can move fast, they can break things, they don't have to get seven people to agree. I'm reading here for my research team. Co founders include Jithin, Anke, and Or and Roda, and then you. Is there four of you? And how do you agree? How'd you split equity on day one? This is a lot of people. Siva (04:06) Yes. Right. Yeah, yeah. Yeah. So at least my style was a bit different even in my previous startup. I start alone, then I start building something, and I have some interns who might probably I might hire to build something. I get to a stage where I am convinced that okay, there is some potential here. And then I start building the team together. So Jimmy and which is Jitin and Anni were introduced to me. Anni and Anni I knew Anni earlier as well. and Jimmy was introduced to me and we started speaking with one another and they came on board in I ran Jan twenty twenty four, almost six months into the system. And and then we had Shreyas and Kush, who also are founding architects who came on board. So this is the code team, five of us, of which four of them can code. Ani is our growth expert. So that is how the whole team came together. At least that's my style. It's not that I need a co founder to start something. You just need an idea. start something and then obviously you can build a team. Nathan Latka (05:28) Take us through the revenue story and then I want to nerd out on product with you. What did you finish at the end of 2024 with in terms of revenue? Do you remember ARR? Siva (05:32) Yes. Yeah. we finished 2024 at zero revenue, right? But this is what happened. 2023 is when we started building something. 2024 is when we started doing some early experiments. And you won't believe during that experiment, around July 2024, we built an AI SDR as an open source agent on our framework. The idea was to show that. Hey, you can have AISDR completely open source. And during that time, Levinex and Artisan were two popular AISDR companies that I think Andres and Horowitz invested 80 million or something in Levinex, right? So my thesis was that okay, why should you pay so much for an AISDR if you could build it as an open source agent on Lyzer framework? And interestingly, we started getting a lot of demand for what we built. So we Nathan Latka (06:09) Yeah. Lemonic. Siva (06:33) Hosted it all. Nathan Latka (06:33) Wait, Steva, real quick, sorry to cut you off. What was the name of the open source project? Siva (06:38) JSON J-A-Z-O-N. JSON AI S DR. If you Google JSON AI S DR, you will find tons of mentions about the AISDR product that we Nathan Latka (06:49) Well, I just I like showing this because I think a lot of people say, Nathan, where are where's the crop of the fastest growing startups coming from? And I just think if you put a scraper together to track GitHub stars on these open source products, that's a great leading indicator if something's got legs or not. But I'm you have to help me out here. I'm sure this is not it because this one do only has one star. So this is probably a copycat. Siva (07:11) no, this is one of the libraries that we launched. I mean, I think no, this is just a shell represent. There's no code here actually. So JSON was part of Lyser Automata. You see Lyser Automata? So JSON was yeah, JSON was one of the rep one of the frameworks or one of the agent templates available within Lyser Automata. This was our core automation framework that we launched, multi-agent automation framework. And JSON was one of the agent instruction set that you can launch within this. Nathan Latka (07:22) Laser automator. So this was the initial business plan basically. This was the initial product. Siva (07:41) Now what exactly was the initial product? And here is what happened. We went from zero to four million in contracted ARR in three months. Just three months. Now, why contracted? Because we told our customers that, very honestly, that hey, we don't know whether it'll work, but sign up with us and we'll work with AWS to even fund the internal the pi the pilot. And AWS has a plan to Fund these pilots. We said we will try and do this for you. And if it works, you convert into a paying customer. So that is why it's a contracted era. They started with a contract. The first three months was covered by AWS, and the rest, nine months, by the customer. This is the business model that allowed us to show the product really well, and and then we were able to convert a lot of customers who came on board. this was the bull run that AISDR companies had, right? And Nathan Latka (08:38) How did you get that initial tranche of like ten, fifteen customers? Did it come from the ones that were like starring the open like the the open source project or somewhere else? Siva (08:46) No, we started sharing this on Twitter and LinkedIn. wherever Levinx was mentioned. I mean it always pays if you go after the the the the number one player and show an alternative option. So we started seeding on Twitter and LinkedIn and Reddit, highlighting that you have an open source version that is available that you could try. You don't have to necessarily spend I think Levinx was back then charging seventy thousand to hundred thousand dollars per year for a plan. So So that is how we started building the initial I would say demand. And then Ani, who's my co-founder and head of growth, he is fantastic. His team was able to then get the SEO in play. And we started getting 400 inbound leads a month during that season, during the season. And we had a 60% demo conversion rate to a contracted ARR customer because the The demo was showing it was showing really well. We had integrations to upper low and all of that. So that is the bull run that we had in 2024. right. And we went to four million ARR, contracted ARR in by September, only in October, we said we are not gonna do this anymore, and we stopped doing it because we realized that AI SDRs don't work. or don't the the promise of AI SDRs. If and and Nathan, you would know this really well. The promise was that it is supposed to book meetings for you. It is supposed to write emails automatically book meetings for you. The moment that promise was broken, customers felt that okay, you know what, this tech is not working. That's exactly what we saw from close quarters. The other SaaS companies were not admitting it, but we saw that okay, you know what, AISDRs can probably do mass emailing, but can it book a meeting for you? That depends on many other things. That depends on aspects like your product maturity, your product market fit, your pricing strategy. There is a lot more conditions that come to play. So we what we did, we gave the open source option to all our customers who had already signed up. And we just discontinued the AISDR to go back to the core product, which is the agent framework itself. Nathan Latka (11:07) So Siva, what is your prediction for all these AI SDR agents? Since you built your own, grew it to four million of revenue, then shut it down. I mean, does Eleven X and Victor and Finn, do they have you know, Artisan, do they eventually have churn problems and shut down, or what's your prediction? Siva (11:19) massive churn problem. Not just I mean in fact one of the AISDR founders and I we did a we did a podcast, 30 minute podcast. So he had built a very good AISDR product, I forgot the name. it was easy to use and even I liked how he had built it. and he built it after us. we were early and he built after us. Long story short, we actually discussed that in detail in a 30 minute conversation on what failed, why it failed. They raised five million dollars capital, they grew really fast, but then they shut it down. They pivoted into a different business. The same happened to us also. Just that we were not a different business. It was kind of an experiment that we did from our out of outside of our core business. But then we realized that it's not gonna work. So long story short to your question, where does AISDS really work? It works if you are trying to just send mass emails. Without expecting it to book meetings for you. If you just want the message to go out to say Fortune 2000 customers. So AISDS can do that. The moment you start expecting AISDS to book meetings for you, that's when it falls free. Nathan Latka (12:33) Even the mass emailing, how are people setting up their email architecture to make sure they have warm inboxes, to make sure they show up in the primary inbox of the Fortune two hundred brand that they're targeting? I mean, even getting that to work is not easy. Siva (12:44) It's not easy. Options are like buy 30 to 40 domains on LEM list or instantly. My favorite is instantly. Exactly. Exactly, right? So buy that, warm it up, then start scheduling, use the API to integrate an agent or cloud directly. So it is it is a process. There is no set infrastructure. but but yeah, eventually you just it's it's probably the lowest cost when it comes to marketing. Nathan Latka (12:51) Warm warmly, instantly, yeah. Siva (13:13) Which is landing in an inbox. If your agent could write a really good opening email, I think you would have caught the attention of the of the ICP. Over a period of time, what we realized was do not try and do the seven email drip campaign stuff, etc. Try two emails, hit the point, give them the links and screenshots to decide. If they don't decide in two emails, they're not gonna decide in seven emails. As simple as that. So we realized that. so your product. So focus more on building the product, trying to showcase it really well, build a good demo video. So do all of that really well. Then just rely on one or two emails to share the value to your IC. Nathan Latka (13:58) So there's something I don't understand here about the story. You have some traction, you're a successful entrepreneur, you shut this four million AR company down in October twenty twenty five, but my research team told me you also at the same time raised a series A led by rocket ship of eight million dollars. Is that accurate? Siva (14:16) I think the year, right? October twenty twenty four is when we shut down the AISDR product. Yes. Which is why. Which is where, right? October Exactly. We went up and shut it down, gave it free because I realized that AISD is not gonna work. We go went back to our drawing board, which is hey, we wanted to be an enterprise agent company, let's do that. And we took three more months to build out the developer studio and Jan twenty twenty five. Nathan Latka (14:21) I thought you said you had zero ARR in in twenty twenty four. You you you you had revenue and then you shut it down. Siva (14:45) Is when we launched the Lyzer Agent Studio, which is a low-code, no code builder platform for developers. So it was a very hard call. I mean, Ani, Anni, who's my co-founder and Ed of Growth, he and I had a lot of conversations because he was like, Siva, we have invested a lot of assets here. Everyone knows what JSON is. Someone types 11x, they're able to see JSON appear organically, should we leave all of it? And I convinced Anni that Anni, I think we should go back to the drawing. Nathan Latka (14:51) I see. Okay. Siva (15:13) Go back to the larger calling, don't get distracted with the AISTR. And that eventually turned out to be a good decision for us. Nathan Latka (15:21) Okay, so new launch in January of twenty twenty five. How'd you get your first ten customers? What were you charging? Siva (15:24) Yes. again, good question. So we started with the $10,000 annual plan. Q1, we've we onboarded 25 customers, mostly startups and SMBs, and we went to 250k in ARR by Q1, right? And then by Q3, we were like we went, we launched a $50,000 plan and a $100,000 plan. By Q3, we went to $650K in ARR. That is when we signed our term sheet with RocketShip, who led our Series A. Accenture Ventures participated. But what happened after that is the growth story, right? Q3, 2025, 650k ARR, and we closed Q2 this season, this 2026 at 12 million ARR. So we grew by 20x in the last nine months. We are projected to get to 30 million minimum by the end of the year. but there is one large government deal, and if that comes in, we'll get to 50 million by end of the year. Nathan Latka (16:23) Well, as we're recording this here in July of twenty twenty six, what you're telling me is you just closed out June and you did a million of MRR in June, annualizes a twelve million dollar run rate. Siva (16:32) Twelve million ARR, yes. Right? All all the three deals that we closed in June, all deals are million dollar ARR each. Three customers. Nathan Latka (16:34) Yep, yep, yep. That's wild. So what is your l don't say their name obviously, but what's your largest customer pay you today? Siva (16:47) it is a federal agency, close to two million dollars. Nathan Latka (16:52) Wild, interesting. And and well, I wanna I'm now I'm curious, but I don't want you to give the customer name away. Can you talk about their use case at all without giving them away? Siva (16:59) Exactly. So so what did we eventually build? Why is Lyzer growing really fast? We are a sovereign full stack agent platform that enterprises could run locally within their own cloud or even on-premise system. That is what we are. So today companies are understanding what OpenA and Anthropic are doing. They are obviously charging you for tokens, but eventually they're able to take all the intelligence away from you. The if you are an insurance firm, the moment you upload your underwriting handbook to OpenAI or Anthropic, it's gone. The intelligence is now with them. They can share the same decision making to every other insurance firm in the world. Your secret sauce just went out. So this is something that enterprises are realizing very quickly. So they are looking at full-stack sovereign platforms like us, which they could deploy within their own cloud or even on-prem systems so that they can. Really own the intelligence. And that is what is resulting in our growth. Lyzer is full stack, which means all the necessary building blocks that you need for a complex agency system are available within our single platform. We kind of replaced 20 different individual technologies under one single homogenous platform, which is why we had Accenture invest in us twice in six months, and they are the world's largest IT services provider. And So so that's and we've been going to customers together. We have KPMG working with us, Deloitte working with us. So many consulting firms and system integrators are both our customers and are also our partners. Nathan Latka (18:43) If I'm watching this interview, what I'm thinking is, I agree with you, Siva. I remember seeing Alex Carpo on CNBC a couple of days ago talking about, you know, going crazy about, you know, his new platform and helping people keep their alpha and their intelligence. You're doing a version of this. But if I'm actually thinking, like as someone watching this as a founder of a $10 million AR company, I'm going, Yeah, I want to keep my alpha and my intelligence with me, but how do I like do I have to If I I think back to like the the rack space days, do I have to have a closet on my business with events a bunch of HP two hundred you know, and H two hundreds running or Blackwells running and then put your application layer on top of it? Like what's the infrastructure under it? How do I get the compute power if I use your application? Siva (19:20) Good question, right? And since you brought up CARP, the last two deals we won, we competed with Palantir to win those two deals. we are the agentic Palantir, if you look at us, right? So Palantir came from the ontology world and now they have an agentic platform. We did the other way around. We started with agentic offering and then we went into data site. So now coming back to your question, it is going to be a combination, Nathan. It is going to be An on-prem appliance which has H200s, H100s, or AMD chips that can have few open source models, even Chinese models like Deep Seek for that matter. So that is your on-prem appliance completely cut off from internet. But you can your CFO office, your RD team, your say biotech team could use it for local intelligence to build agents and automate a lot of work. The second, so I would say probably 20% of a large enterprise workloads will actually have to run. Within the hardware, and we are seeing that already, which is why just last week we launched Lizer Optimus, which is an on-prem appliance that you can buy from us. It can support up to 10,000 concurrent users. You won't find that on the platform yet because we just started pre-orders, pre-orders for our existing clients. So 20% of any enterprise workload will run with local hardware. And about the next 50% will be on one of their hyperscaler platforms like Amazon or Azure or Google. And they will be using models that these hyperscaler platforms provide, or they will be using open source models behind bedrocks of the world, which is semi-sense data sensitive. And the rest 20 percentages where or 30 percentages where they will be open to sharing with OpenAI or Anthropic directly. Because say Fable Five is phenomenal. You don't get that performance with other models, obviously. So there are few coding scenarios, etc., where you need fable five. You don't have a choice. So which is why even OpenAI stopped their fine-tuning API, right? They said, you know what, I I like Cloud. I don't want you to fine-tune any models. I want you to use our best in class model whenever we release. So long story short, we are seeing the split to be a 20, 50, 30 scenario. And companies are very I think they are actively building these strategies. I was speaking to an event management firm. This this particular firm manages events for Taylor Swifts of the World, Justin Bieber's of the world. They are not a bank. They are not an insurance firm. And they said that they want fifty percentage of their agents to run locally with local hardware with H two hundreds. So enterprises are very fast moving towards owning the intelligence. Nathan Latka (21:52) Mm-hmm. Mm-hmm. Mm-hmm. Well, so let me let me ask you a follow up on this because my audience is a combination of tech people that are following everything you're saying and people that are not tech and they might get lost. And that's sort of me. So if they think literally about, okay, I want to own my intelligence, are they going on eBay and buying for forty grand, you know, 10 H two hundreds and then putting in a closet and then putting your app on top of it and running it? Or are they renting are they renting GPU space from a middleman somewhere and then putting your intelligence on top of your application layer on top of it? Siva (22:26) Question, yeah. Three options, right? The option that is on fire right now is the inference providers. Together AI, Fireworks, and Basedon. These companies are inference providers. They allow they give you open source models and you can fine-tune. So you don't have to buy hardware. And these companies ensure data privacy. They are not the big labs like OpenAir Anthropic who use who uses your data to retrain something. Their job is straightforward. They will host an open source model like a Deep Seek, for example, or a GLM or a Quen, and then they give you inference capability. You can fine-tune. Nathan Latka (23:03) These are these are the ones that you t tell me the top who who are the top three in your opinion you think are gonna win the space? Siva (23:09) Number one is together AI, followed by fireworks, followed by Basedon. Nathan Latka (23:15) Base t okay, base ten fireworks. What was the first one? Siva (23:17) Together, together AI. Yeah. They are the number one. I think they just raised eight hundred million yesterday or day before. Nathan Latka (23:21) Got it. When what makes their business model hard? Is or is it easy to copy? You just have to go rent GPU space from the the hyperscalers and then rent it out at a higher rate. Siva (23:33) sometimes they rent, some and mostly they buy. So both, right? So it's all about demand. So two things. One, do you have enough hardware capacity? And number two, do you have a really I would say optimized software stack? The reason why Together Base Tin and Fireworks are good is because their software stack is very strong. take Cursor. Cursor runs their open source models, which is Kimi 2.6. Which they are fine-tuned and they call it as cursor composer, I think. It is powered by BASTIN. So it just because of BASTEN, cursor is able to give you that very fast experience, which is what the customer the developer pays for, right? So, which is why cursor is able to take on the likes of anthropic, like clot code for that matter. So, which is why, if you are an individual, you don't have to buy these Nvidia H200s yourself. You can rent it out from these inference providers. Option number two. Big hyperscalers like AWS, Azure, and Google, they also have some capacity, Core Viv, all the neo cloud, right? Core Viv, Nabs, they also have these options. Then option number three is buying hardware appliance from companies like us, because it's not just the H200s. You need to think about that, it GPU for you. You need to have CPU, which is where the software stack will run. You need cooling. Nathan Latka (24:55) Wait, wait, wait, hold on. While you're explaining this, where can I go on your website to actually see the hardware you're selling? Siva (25:00) No, we we are launching it next week to for public pre orders. Right now we initiated the private pre orders, which is what we are covering right now. We are launching the public pre orders. I can share from my screen here, but I don't know. Nathan Latka (25:03) Mm. I see. Yeah, yeah. Share screen. The more you can show a visual to help my own and wrap their head around what it is, the better. Siva (25:15) yeah. Me shattered. Nathan Latka (25:17) And I'll follow your lead on time. I know we're four minutes over. I have time, but it's up to you. Siva (25:21) sure, no, I have time as well. are you able to see my screen? This is yeah, this is Lyzer Optimus, an on-prem agent factory for enterprises and teams who wants to own their intelligence, right? It comes with it can support up to 10,000 concurrent users, which means even a large bank can adopt this. 95% lower cost compared to a frontier API usage like GPT 5.5 so the world. It just takes 30 minutes to set up and get started. And Nathan Latka (25:24) Yes, yeah. Walk us through this. Siva (25:50) hundred percent on premise. There is not even like you cannot even connect to models via Wi-Fi. It is just LAN. And to I mean, look at our current lineups, right? So we have banks, we have federal organizations, we have biotech firms who are already doing the initial orders with us. Nathan Latka (25:57) What are you gonna sell this for? No, I I see that. But what do you what is this gonna sell for? One of these one of these hardware devices. Siva (26:12) yeah, yeah, yeah. Yeah, it's here. It's here. So there are four configurations that we are launching. Optimus Personal is for the CXOs. Imagine a CEO or a CFO who wants to have a device on their desk. So that is Optimus Personal starts at ten thousand dollars. Optimus Max is one million dollars, it comes with eight small models, two medium models, two large models, one OCR model and one voice model. It can support up to ten thousand users, concurrent users that we're talking about. If you look at Nathan Latka (26:40) Where are you getting all of your NVIDIA H two hundreds from? Everyone's reading depressed today how hard it is to get these chips. Siva (26:46) No, there are a lot of distributors actually. So Ingrams of the World, Ace Computers, there are a lot of distributors who have the stock available. And and interestingly, AMD, the latest AMD chips are better for small and medium models compared to Nvidia Nvidia H200s. So you have options available right now. You won't have you even have Intel chips available that is also very good for small and medium models. You need Nvidia and AMD for large models. But for small and medium, you have more options available. Nathan Latka (27:19) And how does your Optimus, just for my k pro Sumer audience watching, if they want to build their own little mini foundation models at their house, what's the difference between your Optimus personal plan at ten K versus like what most people are doing now, which is like stacking Mac Studios for three K a pop? Siva (27:34) I think so what okay, what what is different with Lyzer is it is not just the hardware that you get. What you also get is the software, right? This is what you get. The whole studio. You Nathan Latka (27:44) Okay. Wait, wait, tell okay, so this is great. I love a visual so just to be clear, we're this is like what a paying customer would see. We're now logged into your platform. Okay. Yep. Siva (27:52) Yes. Right? So let me show this. So this is the studio that you get. You get an end-to-end studio where all the agents that you are building across are registered here. You have an option to create a task agent or a voice agent or even a workflow or even bring third-party agents via proxy agent. You are able to build complex multi-agent systems through the studio. You are able to build even superflows, which is I would call it as a l lot more like NATO style multi-agent system with if else conditions, et cetera, for a lot more items. Like Zapier sort of, but agentic by nature. in fact, six months back, we lost a deal to NATON and that triggered our team to build something that's far superior and enterprise ready for compared to NATON. And since then we've never lost a deal to NATO since we launched Superflow. Nathan Latka (28:29) Like zapier sort of. How many customers are you serving today? Siva (28:50) Today, 32 very large enterprise customers. like I said, the deal values are between two fifty thousand dollars to one million dollars, and one customer obviously closer to two million dollars. And we have eighty thousand builders on the platform. You can go to studio today, create a free account, you get 500 credits, and you can start building agents yourself. So a lot of startups and small businesses are building agents on our platform. We have 80,000. Nathan Latka (29:15) And just again, so they understand why would they build on you versus going directly to Zapier, NADN or all these other agent building tools? Siva (29:22) because Lyzer is much more broader and comprehensive, right? Today, why would a customer running on Microsoft AI Foundry or Gemini Enterprise or Amazon choose Lyzer? Because Lyzer has 50% more agent building blocks or modules compared to what even some of the best or top agent platforms in the market have today. So it is the comprehensiveness of what we bring to the table allows us to obviously help our enterprise customers build. Really complex systems. Like for example, simulation engine. Building an agent is one part, but take it through the simulation engine where it r writes up to 10,000 simulations, runs these simulations on your agent, real-time simulations to see where your agent could fail. And it explains why it fails. And then when you enable hardening, it automatically runs in a recursive loop to improve your agent instructions for tool calling, etc. And it fixes this. So we are talking about three months of work done by a QA engineer delivered in 45 minutes through the Lyzer. Nathan Latka (30:28) I mean this a this is a light version of if you guys nerd out on the early days of the foundation models. I mean, this is effectively neural networks, cue tables with the reward column, closed loop, you know, you know, deep learning and and human and loop training that's happening here, but in sort of an interface that a non engineer would understand. Siva (30:47) yeah, correct. Right. And while Lyza developer studio, this is a developer studio, right? Built for developers. We then wanted to build something for business users, which is why we built architect.new. So this is something that I would encourage the users to try. This is purely built for business users. It sits on top of the developer studio, but you could just go in and like for example, let me show one example here. I was...