A Product Market Fit Show | Startup Podcast for Founders
00:13:59 · Niklas Lindgren
[00:13:59] S0: Exactly and then we're meeting with David, we had lunch, we do almost the same pitch to him. This is the problem, this is how you work, this is what the engineers globally today are producing in terms of schematics, drawings, et cetera. And this is the way they get there. And he said almost the same thing, this is incredible. If we can solve ten percent of this, this will be a generational company. So David at the time was working on Goldman Sachs' algorithm team. So the team that builds the actual algorithms for how Goldman Sachs trades autonomously across the globe. So when Goldman Sachs takes a billion dollar autonomous purchasing decision, that sort of is the team that builds the math behind that and he was part of taking Goldman from place number seventeen to number two behind Morgan Stanley. And then he resigned. This is in, I think this is in 2023, or something. Early 2023, and everyone asked him, what are you doing? Because no one resigns really from that type of thing, right? So everyone thought that he was going to Morgan Stanley in New York or in London. And he said, no, no, no, I'm just going to sort of figure out what this GPT-3 is about and figure out what this new technology shift is all about. What he did was that he spent a year building his own LLM from CUDA level up to the chat interface. Got it to work, got approached by a lot of VCs to go fund this into an actual company in a lab and he rejected that, and took on a consultancy job instead. He was sort of very non-VC in the beginning. Now, obviously, three rounds after, he's sort of all happy now. But in the beginning, he was sort of against the whole VC trend of this industry and took on a consulting job. And in that time, we approached him.
00:15:56 · Pablo Srugo
[00:15:56] S2: I think you're making the point of just how important the team is. You've got all these pieces of the puzzle that you're adding on. The four co founders, which is not normal but if everybody's playing a specific role, they have a clear reason to be there. It can drastically affect the odds of success. We're actually going to talk later about, because you carried that through to hiring not just the first four co founders, but the first thirty or forty employees. You had a hiring principles manual. You went outbound. So we're going to talk deeply about how you did that. But I would love to go, you had an insane 2025, at least according to Crunchbase. You raised $3 million, let's say mid '25, raised $20 million near the end of '25. So obviously a lot of stuff was happening. Then January 2026, we know it was a PMF moment where you signed a million and a half in a month, right? So I'd love to go as deeply as possible into everything that came before that. Everybody wants to end up in that place where they just have that banner year. The fundraising is part of it, but definitely the signing that muc., A million or two million, you sign in a short time frame. A month is exceptional, but even if people could do it in a year. I think a lot of founders will be happy. But all of that has to do with what comes before it. So you're talking about, mid 2024, you're doing some work. Let's say on the side on this idea. What is that work? What are you doing? What are you spending your time on?
00:17:11 · Niklas Lindgren, Pablo Srugo
[00:17:11] S0: Oh, wow, we did a lot of things and I think that is the overall answer to it. You need to, like we tried a lot of things. What we really wanted to get to is, how are we building this product? Because there is an industry standard where everyone works on and we didn't want to replicate whatever went on that. Because everyone just hated it. So we wanted to get into something new. [00:17:35] S2: What is that standard, by the way? Just maybe not super deep, a little bit of context? [00:17:38] S0: That standard is everyone works on one platform called Revit. Like lawyers work in Word, these engineers work in a platform called Revit. Lawyers don't hate Word, but there's a lot of engineers across the globe that hate Revit, because it's so legacy. It was, as I said, built in '97. It's not back compatible with Word. It's not a modern software. We wanted to understand early on, what are we building? How should a five star user experience be? And being just a four man team, none of us being the actual ICP ourselves, we needed to get out and speak with users. So that's what we did early on. We wanted to speak to as many potential users as possible and we did this on. I think most of our conversations early on were on Reddit, with DMs on Reddit, in MEP engineering threads. [00:18:31] S2: And you got them on calls or are you mainly, literally mainly online?
00:18:35 · Niklas Lindgren
[00:18:35] S0: We got some of them on calls, and then we got some of them just chatting back and forth, asking them questions to verify different types of ideas or concepts or things like that. We did that really early on '25, like spring of '25, and then we had something that you could look at in the summer of 25. Like an actual product, right? Very prototype, but still something like a user interface you could actually look at and once we started showing that to users. Like, how it could work. It didn't have like a fraction of the feature set that we have today or a fraction of sort of the features that you needed in order to launch it. But they understood the concept of this and it was almost like a dream coming true for them. Because they spend so much time on tasks now that are just bullshit tasks. They have to move data from this program to this program and it's almost a copy paste type of task. And they spend so much time doing that simply because the systems are not interconnected with each other, and don't talk to each other. To give you an understanding of this, like if you're building a hospital today. You have engineers that are manually placing each and every unit that. When you walk into the hospital, each and every unit that you see has been placed in a 3D model before going in there. So every individual needs to place each and every receptacle, each and every lighting fixture, each and every lighting switch inside this hospital. And these are tens of thousands of units. And then once that is done, they need to calculate how many receptacles do I have and then they need to ingest that into an Excel file to then calculate like how much load do I have on this panel. And once I know how much load I have on this panel, I need to start calculating in another software how much metal is needed inside these cables and it's levels of complexity. And so once I saw that, oh shit, we can have an agentic experience where this is generating the design and I almost have a vibe coding experience. Where I can set my playbooks, I can set my constraints and I can give a context window to. And have Endra just build this system for me instead, and then I can start reviewing them.
00:20:47 · Pablo Srugo, Niklas Lindgren
[00:20:47] S2: In this case, add the fixtures for me. Like you tell it more or less, hey, this is the number or this is the spacing or these are the constraints and it kind of just does it. [00:20:54] S0: Exactly like that. [00:20:55] S2: OK. [00:20:56] S0: When engineers are starting to say like, holy shit. I know because they gave us their models that they previously completed. So once you saw like, once I ingest this into Endra, it's not perfect but it comes to eighty, ninety percent of whatever I did before and I spent four hundred hours doing this project. Endra created this for me in under ten minutes. [00:21:15] S2: This is an interesting point to draw on because a lot of the AI output is eighty, ninety percent. In some cases, eighty nine percent is almost as good as zero. It saves you no time because then you have to review, you double check, blah blah blah, and you're like, oh my god. But in this case, the eighty to ninety percent really was eighty to ninety percent. I just, I don't know how that plays out, but maybe you can speak a little bit about that.
00:21:15 · Pablo Srugo, Niklas Lindgren
[00:21:15] S2: This is an interesting point to draw on because a lot of the AI output is eighty, ninety percent. In some cases, eighty nine percent is almost as good as zero. It saves you no time because then you have to review, you double check, blah blah blah, and you're like, oh my god. But in this case, the eighty to ninety percent really was eighty to ninety percent. I just, I don't know how that plays out, but maybe you can speak a little bit about that. [00:21:34] S0: Yeah, definitely and the reason why we're saying eighty, ninety percent is because it goes back to how you design buildings. That's an iterative process. So eighty, ninety percent is one deliverable and it's fine that it's not a hundred percent in that stage. But that's the full deliverable that you charge your customer for and then as design develops, the architect comes up with a new building layout. Which has another room in the left wing or whatever. You sort of iterate, iterate, iterate. There's a lot of time being spent on both, but sort of the greenfield design is usually around eighty percent of then you iterate as design develops into a more and more detailed design. And that, also consumes a whole lot of time and that goes much, much faster once you do it in Endra. Because you don't need to do that in five, six different point solutions.