AI-Powered Coding & Space Tugboats? Inside the Future of Software & Satellites! | E2099

Today's show: Alex interviews two incredible founders from the TWIST 500! First, Replit CEO Amjad Masad breaks down how AI-powered coding is changing software development and what the future of agents looks like. Then, Starfish Space co-founder Austin Link explains how they're building a satellite towing service to extend missions and clean up orbit. Two deep dives into the future of tech and space. Timestamps: (0:00) Alex kicks off the episode! (0:53) Discussing Replit's capabilities and introduction of CEO Amjad Masad (4:17) Replit's evolution and turning down a billion-dollar acquisition offer (9:48) Vapi - Go to https://vapi.ai/twist and get 1000 minutes free per month - for life. (11:42) Impact of LLMs and AI agents (17:13) Choosing Anthropic's Claude for Replit (19:46) Northwest Registered Agent. Form your entire business identity in just 10 clicks and 10 minutes. Get more privacy, more options, and more done—visit http://northwestregisteredagent.com/twist today! (21:29) Anthropic success and closed source AI implications (24:37) Replit market response and economics of AI models (28:14) Future of Replit's AI tools and plans (33:41) LinkedIn Jobs - Post your first job for free at https://www.linkedin.com/twist (35:12) Introduction to Starfish Space and its mission (38:50) Core missions and challenges of the Otter satellite (47:29) First official mission with Otter and key partnerships (50:57) Economics and ownership of satellite services (53:11) Key components of the Otter project (57:19) Startups in the space economy and comparison with Orbit Fab (59:08) Starfish Space hiring push and final thoughts Subscribe to the TWiST500 newsletter: https://ticker.thisweekinstartups.comCheck out the TWIST500: https://www.twist500.com Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp Links from the show: Check out Starfish Space: https://www.starfishspace.com/ Check out Replit: https://replit.com/ Follow Amjad: X: https://x.com/amasad LinkedIn: https://www.linkedin.com/in/amjadmasad/ Follow Austin Link: X: https://x.com/Austin LinkLinkedIn: https://www.linkedin.com/in/austin-link-space/ Follow Alex: X: https://x.com/alex LinkedIn: ⁠https://www.linkedin.com/in/alexwilhelm Thank you to our partners: (9:48) Vapi - Go to https://vapi.ai/twist and get 1000 minutes free per month - for life. (19:46) Northwest Registered Agent. Form your entire business identity in just 10 clicks and 10 minutes. Get more privacy, more options, and more done—visit http://northwestregisteredagent.com/twist today! (33:41) LinkedIn Jobs - Post your first job for free at https://www.linkedin.com/twist All rights for the video shared and linked below belong to the original copyright holder, and we are using the footage under the principle of fair use. * Great TWIST interviews: Will Guidara, Eoghan McCabe, Steve Huffman, Brian Chesky, Bob Moesta, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland * Check out Jason's suite of newsletters: https://substack.com/@calacanis * Follow TWiST: Twitter: https://twitter.com/TWiStartups YouTube: https://www.youtube.com/thisweekin Instagram: https://www.instagram.com/thisweekinstartups TikTok: https://www.tiktok.com/@thisweekinstartups Substack: https://twistartups.substack.com * Subscribe to the Founder University Podcast: https://www.youtube.com/@founderuniversity1916

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  Hey everybody, welcome back to Twist, a special Tuesday episode. And today I'm bringing you not one, but two amazing conversations with some of the best founders out there in charge and running some of the most interesting and innovative companies in the world. Let's go.

• 0:14Speaker 01

  This week in startups is brought to you by Vapi. At real time, AI-powered voice conversations to your apps or business in minutes, not months. Go to vapi.ai/twist and get 1,000 minutes free per month for life. LinkedIn Jobs. A business is only as strong as its people, and every hire matters. Go to LinkedIn.com/twist to post your first job for free. Terms and conditions apply. And Northwest Registered Agent. Starting your business should be simple. With Northwest Registered Agent, you can form your entire business identity in just 10 clicks in 10 minutes. From LLCs to trademarks, domains to custom websites, they've got you covered. Get more privacy, more options, and more done. Visit NorthwestRegisteredAgent.com/twist today.

• 0:53Speaker 00

  Hey, everybody. Welcome back to Twist. Now, I haven't written code really since high school, and even then, my C++ was nothing to write home about. But I live a very, very software-driven life. That's why it's exciting that I think there are a lot of startups out there that are doing amazing things to help folks like me build more. And one of those companies, I'm sure you've heard of it, is called Replit. Now, I've gone out and played with its service, and I can say with confidence that it is super darn cool. And personally, my thesis is that as more folks can build, more folks will build. Some people out there think that AI-predicated coding will wind up generating spaghetti code bases replete with technical debt that are just toys at best. So I wanted to get to the bottom of

• 1:36Speaker 00

  that. So I've asked the CEO of Replit, Amjad Masad, to come on the show, tell us what they've built and where we are going. So please welcome to the show, the CEO of Replit. It's Amjad. Hey, how you doing?

• 1:46Speaker 02

  Thank you, Alex. I'm doing well.

• 1:47Speaker 00

  Let's start with what Replit is today, and then I want to go back in time. But from you, what have you built and how do people use it?

• 1:53Speaker 02

  Yeah, so today, most people come to the site and the first thing they do is enter a prompt of what they want to build. Kind of similar to ChatGPT, Perplexity, a lot of the AI-first products is that, you know, you just have a prompt box. And then we unfold the UI into this sort of agentic interface where you have an agent or chatbot talking to you about your project. It will present a plan to you, be like, All right, you know, based on your prompt, I can do X, Y, and Z. I can add this database. I can use this technology. You don't have to know any of that, but some people are interested and curious that you're going to use Postgres and all of that stuff. And then if, if it, you know, doesn't sound good to you, you can sort of ask your questions, ask it to change the plan.

• 2:37Speaker 02

  Once the plan looks good, you can hit approve and it will start recording. You can watch the agent code in real time. You can see the code kind of flow through. It feels like a little bit like the Matrix. And you can watch it like do everything. You can watch it install packages. You can watch it provision the database. You can watch it, you know, run the project and, and once it gets to a place where the project is runnable and it'll run it. I'll take a screenshot, make sure it's actually running and then present the interface to you. So it does look like you're working with almost like a person who's trying to present the best possible product to you. And then it'll ask you like, how does it look? Like, can you test this, this or that? What do you want me to work on next?

• 3:20Speaker 02

  So we try to like make the UX and the sort of AI, like the, the way the AI talk to you to be like a little more human.

• 3:29Speaker 00

  Sure.

• 3:29Speaker 02

  And then from there you can iterate and at some point when it looks like the objective has been complete, it will, it will prompt you to deploy. Hey, do you want to use our deployment service to put this, put this in the cloud in a way that's scalable? The way that you can ship it, you know, give it to your friends, tweet it out. And even if you get a million users, you know, Replit will be able to handle that.

• 3:51Speaker 00

  For I presume a tasty fee, you're not going to host my million user application for, for gratis, are you?

• 3:56Speaker 02

  No, we've talked for that.

• 3:58Speaker 00

  But it's, it's pretty cool. So essentially you can show up, type in, I want to build, my test was, please make an app that makes email fun for me because I'm chronically behind on email. But I was shocked at how cool it is that I can go from prompt all the way to live web app without writing any code myself whatsoever, just simply talking. What was interesting, I'm shocked is that prepping for our chat, I realized that this version of Replit came out kind of late last year. And the company founded back in 2016 was doing quite a lot up until that point. So can you just walk us through where you guys started back in 16? And then what changed along the way technologically that allows you to build this new and seemingly explosively popular version of Replit?

• 4:39Speaker 02

  The mission that we're on, and I've been on even pre-Replit, uh, cause I worked on Default for Tools and open source at Facebook. Was how do you make programming suck less? How do you make it more accessible? And so the thesis there is like more people would want to do software. I saw it in school. A lot of people wanted to code, want to create applications, but there was only a subset of us that had the tolerance to actually go through the pain of setting up a development environment. And I felt like that was like needless. It's almost like hazing. It's like need this amount of pain to take on in order to make something that's fundamentally creative, which is making software, I think is an act of creation.

• 5:22Speaker 02

  I think creative expression in many ways and it can enhance your life. That sort of driving motivation has been with me for, you know, 20 years now. You know, the original version of Replit actually came even before 2016. It was like 2011. It was an open source project and it was like a way to, you know, just run code in different languages. We made a startup, my co-founder and I, in 2016. The mission was to make programming more accessible. The way we approached it, we didn't have AI tools, but the way we approached it was like, okay, we're going to automate everything up until to like the point of writing code.

• 6:00Speaker 00

  Okay. So essentially, if I wanted to go out and write code up until the point when I start typing in STD colon, colon, see out or whatever the hell it was back.

• 6:07Speaker 02

  We'll take care of all that.

• 6:08Speaker 00

  Everything else, that's great because when I was learning to code C plus plus, that was not as easy and it actually took some work and reading those big, thick books you used to buy, it's terrible.

• 6:17Speaker 02

  Yes, yes, it's a horrible experience. And so that, that in itself was very, very exciting to people, especially hobbyists, students, artists, scientists, people who weren't coding every day. And some developers really liked it as well. So we grew based on that, like, you know, that was like the first kind of online IDE experience. And so we added collaboration, multiply collaboration, and we also started selling to schools. And so we have some education tools around it and the business was, was doing all right, especially on user growth. We were growing really, really fast. You know, we're growing faster than get up at the, at the time in terms of like people coding.

• 6:56Speaker 00

  So much so, actually I'm glad that it was in 2019, the company actually turned down an acquisition offer.

• 7:01Speaker 02

  Yeah, yeah, yeah.

• 7:03Speaker 00

  For, and it was for a dollar amount. I'm going to let you say it, but it was, it was, uh, it was reported to be quite a large dollar amount.

• 7:09Speaker 02

  Yeah, for, for almost a billion dollars.

• 7:11Speaker 00

  Yeah.

• 7:12Speaker 02

  And sometimes.

• 7:15Speaker 02

  You know, I used to credit when things are painful and really dark.

• 7:21Speaker 02

  But I felt like what we're doing was quite unique, was not something that people are focused on, which is like, the market is not obvious to actually make programming for non-programmers. And so I thought that the potential for this company could be, it multi-billion dollars could be, could be, you know, up to a hundred billion dollars.

• 7:42Speaker 00

  That's a really interesting point. Why is the market not obvious to bring coding to non-developers? Because to me, it feels like such an empowering offering that it makes intuitive since as a non-developer. But I'm curious if from a developer perspective, it's less reasonable?

• 7:57Speaker 02

  Yes. So the dogma in the sort of industry is that you need the tools to be, you know, maximally customizable. The developer needs to have like full control over the tools. You know, online IDEs never worked. Everyone wants, wants to code locally. VCs never seen, you know, a startup that, that, you know, there's not no calm for, hey, we're going to create a new developer market that is actually not a developer, right? And there's this buzzword over time that it's like citizen dev, a citizen developer. And there, there's some, there's some attempts at it, but there's no big company that you can point at and just say, okay,

• 8:38Speaker 02

  this is a citizen developer company. Now over time, like low code, no code became a thing.

• 8:43Speaker 00

  Absolutely.

• 8:44Speaker 02

  And I think that's adjacent, but that's not exactly what we're doing. Because with low code, no code, you're, you're building on sort of their IP, right? Like you're not building an open source and you're building an, their own custom infrastructure.

• 9:00Speaker 00

  Yeah.

• 9:00Speaker 02

  Large language models are not trained on retools infrastructure. They're trained on open source infrastructure.

• 9:08Speaker 00

  Yeah.

• 9:08Speaker 02

  They're trained on Python, they're trained on Postgres. So LLMs are naturally going to be able to code using standards and open source. So that's, I think, a fundamental difference between what we're doing now and what you would call no code, low code.

• 9:22Speaker 00

  That makes a lot of sense to me. In my experience, no code development ended up being essentially just drag and drop.

• 9:28Speaker 02

  Yes.

• 9:28Speaker 00

  Connections between different data sources and places that would ingest them, which was itself an improvement over not having that capability.

• 9:35Speaker 02

  Of course.

• 9:36Speaker 00

  But it was miles away from. Hi, Replet. I want to build, I don't know, today, we're going to build a calendar app for squirrels, or whatever it is that I that I come up with.

• 9:45Jason Calacanis?

  Yeah.

• 9:46Speaker 00

  Hey, maybe squirrels are busy. You don't know.

• 9:47Jason Calacanis?

  Yeah, yeah, who knows? 2025 is going to be the year that voice AI really breaks out. Voice AI is going to change everything, and this technology is perfect for building a chatbot or automating customer calls and basically creating these next gen voice experiences. But building with these tools is so difficult. It's just a bear. It's going to take your developers months to build and test anything even basic unless you use VAPI. V A P I. This is the fastest and easiest way to integrate voice into your products. It's just the most simple API you've ever used. It helps you build, test, and deploy voice agents in just minutes.

• 10:28Jason Calacanis?

  And you can create ultra-realistic AI voices and customizable personalities. I like that part. All with real time response speeds. You know what'd be great? Maybe I just demo it. Okay? Listen to this.

• 10:41Speaker 04

  Hey, Jason, it's great to be on your podcast.

• 10:43Jason Calacanis?

  It seems like VAPI has figured out how to be very low latency. Maybe you could talk a little bit about why that's important.

• 10:49Speaker 04

  For sure. Low latency is crucial for a smooth user experience, right? It makes conversations feel more natural and engaging. People expect real time interactions, especially with the voice assistants. At VAPI, we've optimized everything to keep that response time around one second. It's like having a chat with a friend, not waiting for a robot to catch up.

• 11:07Jason Calacanis?

  Can you do other voices and personalities?

• 11:10Speaker 04

  Right now, I'm rocking my CEO's voice, but with VAPI, developers can bring their own voice models. So if they want to add some walk and flair or any other personality, they totally can. I mean, who wouldn't want a little drama in their AI conversations? Just think of me as your friendly AI buddy, keeping it real.

• 11:27Jason Calacanis?

  If you want to give it a spin yourself, and trust me, you should, just head over vapi.ai/twist and Twist listeners will get 1,000 free minutes per month for life. That's F R E E. Free. What an amazing price. Again, that's V A P I dot A I slash twist.

• 11:42Speaker 00

  Okay, so there's a big thing coming up in this timeline, though, which is suddenly LLMs arrive, take off, and then we have quite literally what is called the ChatGPT moment. And that seemed to kick off an explosion in the development of LLMs, eventually leading to Rapid launching its agent feature.

• 11:57Speaker 02

  Yeah, I had the feeling that AI is going to be impactful on code for a long time because I worked at Code Academy. That was my first job in the US.

• 12:04Speaker 00

  No way, I've played with that, it's fun.

• 12:05Speaker 02

  Yeah, a lot, you know, 50 million people learn how to code in Code Academy. Actually, a lot of the people that I hire now, they're like, oh, you know, I learned to code in Code Academy, it's a

• 12:15Speaker 00

  Full circle.

• 12:16Speaker 02

  For a long time, I felt like applying deep learning to code was going to be, was going to be really impactful. So when GPT-2 came out, I felt like, okay, we're getting into air. So language processing is getting really powerful. So we're actually, we were the, were the first startup to train a model, that an open source model. We actually did it in, in 2022, even before the sort of ChatGPT moment.

• 12:36Speaker 00

  Yeah.

• 12:36Speaker 02

  So when ChatGPT happened, we came out this product called Ghost Rider, and it was kind of a co-pilot-like product, but it was, it wasn't satisfying because our customers' coding is still hard for them, uh, and boring and not really something they're excited about. And so when you have an autocomplete system, it's not really, it's not really solving the problem. So we started imagining a world where you don't even have to code. Not only that, you don't even have to set up a database. You don't even have to set up a cloud environment. You don't even have to learn how to deploy. And LLMs, you know, there was this idea of agents, right? Uh, LLMs were really bad at it, but we started

• 13:18Speaker 02

  getting sort of tool calling and like, you know, in 2023, this idea that the LLM can call a function or call a tool. And that's like the, sort of the beginning of what you might call an agent, an agent that could make actions.

• 13:32Speaker 00

  Yes.

• 13:33Speaker 02

  Do actions, uh, on behalf of, uh, humans based on a request.

• 13:37Speaker 00

  Should we define agent here? I'm John, I, I think you're actually doing a pretty good job, which is that an agent is an LLM that can then go out and do something for you versus just spitting back information to you. So ChatGPT, lots of fun, great model, cool technology, not an agent, but

• 13:51Speaker 02

  That's right.

• 13:52Speaker 00

  Agents is probably the biggest buzzword in technology this year, but it seems like you guys were early.

• 13:57Speaker 02

  Yes, and and it's it's kind of getting misused in similar ways like the cloud or or any any sort of buzzword. The fundamental difference between, say, a chatbot and an agent is that when you send a message to a chatbot, it just respond immediately. There's this request response, sort of like HTTP, right? They go on website, kind of return something.

• 14:16Speaker 00

  Yeah.

• 14:16Speaker 02

  Whereas with an agent, when you send a request, it can do arbitrary number of actions before it it come back. And arbitrary is important because there are also chatbots that can do an action or two before they respond. But I think my definition of agent is a little more strict than that and that the agent has the autonomy to know how long it needs to go before it returns to you. Think about deep research, for example. Deep research this new product from OpenAI.

• 14:45Speaker 00

  Yeah.

• 14:45Speaker 02

  You send a query in and it can search as much as it needs in order to get a grasp on the query that that you sent it, right? So there needs to be a sense of unboundedness

• 14:56Speaker 00

  Yeah.

• 14:56Speaker 02

  to the action the agent is taking.

• 14:58Speaker 00

  Tell me why that matters in a development context. I I think I know what the answer here is, but I just want to spell it out for the audience because having no time bound doesn't play, L L M's can think more, great, but that's more expensive. So there is a a potential downside here if it's a misuse. So what's the advantage that makes that worth it?

• 15:15Speaker 02

  Uh, debugging. Uh, so you know, with with a raplet, you know, it'll write some code. It'll rerun it and then, you know, I'll take a screenshot, it'll look at the logs and if there's a problem,

• 15:26Speaker 00

  Mm hmm.

• 15:26Speaker 02

  it will continue working. It's not going to yield back to you and talk to you if there's an error. There's a lot of coding AI products right now that that don't really do that and and you have to kind of tell it fix that error. With a raplet, most of the time, I mean, sometimes we mess it up, but most of the time or aspirationally, what we want to do is like, we only want to present a working application to you. There might be bugs, but they're like, you know, they're logical bugs, business logic bugs, but

• 15:51Speaker 00

  Right.

• 15:51Speaker 02

  we don't want to present a crashed application uh to you. And so you need the agent to be able to make arbitrary number of actions to debug and that means that it writes a log statement. It kind of goes into the code and says, oh, I want to see what this variable is. It runs that. Maybe it goes to the web and does a search. Maybe it does, uh, it writes a sort of a small debugging utility. And then after all that'll fix the code and then when it runs, it presents it back to you.

• 16:23Speaker 00

  To put this into concrete terms, I've been scrolling through my chat log with Revlit and I found an example of this. I was working on my cool email app and the Revlit agent said, hey, quote, I see we need to fix some TypeScript errors and authentication issues before we can fully set up the database. Let me address those issues first and then let me fix remaining TypeScript errors. It's so cool. I've never had someone to hold my hand while also pushing me from the back. It feels like it's doing everything at once.

• 16:48Speaker 02

  Yeah, exactly right. Like, like in this case, it wrote code. It tried to compile it. It didn't compile and then it's going to edit it, right? It's not going to show it to you. It's sort of like a, if you ask an engineer on your team to go get some tasks done, unless they're really bad, they're not going to give you code that doesn't compile.

• 17:08Speaker 00

  Right, right. Although, well, if they want to stay employed, I mean, given the job market.

• 17:12Speaker 02

  Exactly right, yeah.

• 17:13Speaker 00

  Okay, but going back to the, the agent and you guys training your own agent. I know you guys were working on that, but I also know that a company called Anthropic came out with this beloved model called Claude Sonnet 3.5.

• 17:24Speaker 02

  That's right.

• 17:25Speaker 00

  Now they have 3.7 out and code with Claude and so forth, but that model was state of the art for a long time for coding work. And you guys ended up choosing that over your own in-house model?

• 17:36Speaker 02

  Yes. In 2020, late 2023, I actually gave a TED Talk saying that agents are coming. And here's what they will look like. Actually, even surprising to myself, you know, I got a lot of those predictions, right? Even in our own product, I just had a feeling that we're almost there. Like the models are, are able to reason a little better, able to do these tool calling better. So we started building the agent and we just needed the most powerful model possible.

• 18:05Speaker 00

  Right.

• 18:06Speaker 02

  So at the time, yeah, we ditched our own models. We didn't have hundreds of millions of dollars to train these really large models.

• 18:12Speaker 00

  Wait, and Jason Horowitz invested in your company. Surely you have access to hundreds of million, what? Mark?

• 18:17Speaker 02

  There's a lot of, a ton of. Companies that just lit up billions of dollars on fire trying to do that. And so our approach has been, okay, can we do it? When we were training models, we actually spent less than a million dollars doing that. And our feeling is that, okay, like, you know, someone like, you know, there's going to be these open source models that we're going to be able to fine tune. And that's, that's the case today with deep seek and Laman and things like that. So, so we felt like our business doesn't need to be this like large cap act sort of business.

• 18:45Speaker 00

  That makes a lot of sense.

• 18:47Speaker 02

  With agents, we started building it. We're using all sorts of model. We're trying, trying all of them. It was kind of crappy. It was kind of like, but you, you saw sort of a glimpse of what's possible. Like say, you know, January, February, 2024.

• 19:02Speaker 00

  Uh huh.

• 19:02Speaker 02

  But like the autonomy aspect was, was, was really bad and it wasn't working very well. We were really ambitious at the time. So when Claude, uh, Sonic came out and, uh, in June, we sort of like saw online, everyone's talking about its coding ability. It was like, oh, I don't know. I mean, Anthropic, yeah, up until that point, didn't really do anything with coding. That wasn't interesting. And then we tried it. We plugged it into our agent framework. And although we haven't really optimized it for it, it was performing way better than the alternatives. It was clear that they've done something to make it a much better coder. And by the way, no one has caught up yet. I don't know what the magic they've done, but I mean, other companies are trying to catch up, but they haven't yet.

• 19:48Jason Calacanis?

  Hey founders, you want to build the next great billion dollar business, right? Well, you're gonna need two things, a killer idea and a properly set up company. I can't help you pick the idea, but I can help you get your business started the right way with Northwest Registered Agent. With Northwest Registered Agent, you can form your business for just $39 plus state fees. One of the most affordable options out there. No hidden fees, no upsells, just a simple, straightforward process. And here's how it works. In 10 clicks and 10 minutes, your LLC is officially formed and ready to operate. Northwest handles all the paperwork, fast, accurate and hassle free. So you can focus on building your business, not navigating all those crazy. Crazy legal firms, thousands of entrepreneurs trust Northwest Registered Agent

• 20:33Jason Calacanis?

  because they make business formation affordable, efficient, and stress-free. And with their expert team standing by, you'll always have the support you need. So here's your call to action. Don't let paperwork hold you back from your entrepreneurial dreams. Get started today at northwestregisteredagent.com/twist. That's northwestregisteredagent.com/twist. For just $39 plus state fees, your business can be up and running in no time at all. Get more value, more convenience, and more peace of mind only with Northwest Registered Agent.

• 21:29Speaker 00

  Anthropic's success with 3.5 Sonnet to me is the best argument that closed source AI does have the chance at having an edge on open source AI over even a three to five year timeframe. Otherwise, I would be probably more bullish on the DeepSeek Llamas of the world than closed source. But Anthropic does have taste, I think.

• 21:48Speaker 02

  Taste, exactly. Taste, I mean, you absolutely nailed it. Like even the UI that it generates versus the open AI, it's actually tasteful. It looks good.

• 22:01Speaker 00

  So you guys see this in June, you plug it in, even unoptimized, it's impressive. How quickly did it take you guys to go from that moment to, all right, we can get this out to the market? We're going to bring it to the masses.

• 22:11Speaker 02

  It was until September. You know, I'm a big fan of this model of you sort of like set a time box for yourself. And you get to that point, even if you feel the product's not ready, you gotta put something out, right?

• 22:24Speaker 00

  Ah, a deadline. Yep.

• 22:25Speaker 02

  And non-negotiable deadline. Because you can always kind of push those things out. So we were like, okay, we're going to just put it in early access. We're going to all need put it for paid customers. And we're going to put a video just to show the vision and how it works and things like that. A lot of people were just like really surprised by how good it is. Like it did something no one else has done. It did something that maybe Devon has talked about, but no one else had had put it out into the market. They put out the vision for a autonomous software engineer back in early 2024. But up until that point, there was no product on the market where you can experience agentic coding. So Replica came out, it was the first September 2024.

• 23:10Speaker 02

  It was the first product that you can try on the market that is agentic. And so a lot of people tried it and it went viral. A lot of people really liked it. A lot of people made videos about it. But admittedly, I kind of felt bad about it because I didn't feel like the product was ready for for it to go this viral. Like, you know, getting millions of views on Twitter and everything because I felt like it really wasn't ready. And so we scrambled, we're like, okay, you know, war room time, we're just going to go heads down. Although we've been working really hard, we're just going to work 10x harder and make the product a lot better over the coming months. And literally day over day, it was improving and and getting better and better. And then we exited the beta period in December.

• 23:54Speaker 02

  So it really hasn't been, you know, out of beta only for for two months and kind of the delta. And we measure the success rate and the success rate for us is when you click deploy. So Alex, if you like the app that you built, you're willing to pay some extra dollars to actually deploy it. It means the agent was successful. It's such a like.

• 24:14Speaker 00

  Not only successful, very successful, because I think a lot of people, I mean, frankly, like myself, love to tinker, love to play with things, love to test. But if I finish up with something in my experiment that's so good that I'm like, well, I have to pay for this so I can put it up. That's a ringing endorsement of the product.

• 24:29Speaker 02

  That's right. And that number, the percentage of the number of people that are deploying is our quality bar and that keeps going up.

• 24:37Speaker 00

  Okay, so take the beta tag off in December. It's been a couple of months. It was already going semi-viral beforehand. How has market response been since the beta tag was removed?

• 24:48Speaker 02

  So it changed our company completely. You know, like you said, we went from sort of this coding company to mostly a prompt company. Right now, seven, five percent of our customers never write out a single line of code. By the way, you know, what you just showed us is still there's a full IDE underneath it. So you can go in and look at the code and edit it and all of that. You know, we go from a hundred percent of people coding to only 25% of the customers coding.

• 25:12Speaker 00

  That's crazy.

• 25:13Speaker 02

  Just changes the business, right? Our revenue 10xed. And what the most exciting thing about this is not just that, you know, we're onboarding all these customers and users. It's that they end up using more and more and more over time.

• 25:28Speaker 00

  More of Replit, more of the agent, more calls.

• 25:31Speaker 02

  More of everything, right? Like, you know, more of deployments, more of the agent, more of the people who get into it. So in like financial terms, the net dollar retention is going up over time. Meaning like for every customer, we're continuously expanding the spend. Uh, no, ideally the spend is not regrettable. Musa Lab is not, is, you know, sometimes the agent is a little boneheaded. People don't like that, that experience. It was sort of like the same thing with a freelancer or whatever, like sometimes it'll make mistakes and we continue to sort of improve it. Another thing, we released the mobile app, released the mobile app.

• 26:09Speaker 02

  You should check that out too. We released a mobile app where you can just like chat with the agent and you can see the app kind of working in your phone.

• 26:17Speaker 00

  If you're using Anthropic's model, does that make your economics tough? Or is that a cost that works into the product in such a way that I can think of it financially? From a very high boring perspective as kind of like SaaS?

• 26:30Speaker 02

  So a lot of companies in the AI coding space are massively subsidizing the tokens.

• 26:37Speaker 02

  From Anthropic, they're like a mere sort of wrapper over Anthropic, perhaps even losing money on the Anthropic spend. Replit is like actually more expensive than than many products on the on the market, if not most. And the reason is, is like, one, we want to provide this agentic experience where it could take arbitrary amount of time to to get there. So we're going to provide a quality, premium experience. But two, we also have this cloud offering. No one has the both sides of it. The AI, you can't really have a lot of margin on right now because otherwise you're you're, you know, you're going to be, you know, very, very expensive, unaffordable. But, you know, cloud businesses are kind of clear what you do and and so, you know, we we make margin on that as well. So so the business picture is is healthy right now.

• 27:22Speaker 02

  It's a concern for the industry as a whole, like maybe the foundation models are the only people are making money and none of the app layer are making money.

• 27:30Speaker 00

  It's astounding to me how every three months, the entire perspective shifts from there's no value in the app layer. It's all in the models to wow, the models are worthless, all the value is in the app layer. It just swings. Basically, it's is OpenAI's model the best or is Llama the best right now?

• 27:43Speaker 02

  Yeah, every few months.

• 27:45Speaker 00

  I mean, what was really funny is you were talking about how like, you know, back in January, February of two thousand twenty four, I'm like, yes, but in AI time, that's ten years ago.

• 27:53Speaker 02

  That's right.

• 27:53Speaker 00

  So I mean, we this is a compressed cycle to the point to which

• 27:56Speaker 02

  Hundred percent.

• 27:57Speaker 00

  It's tough to keep up, even though I get paid literally to keep up. But so I might be kind of over my skis here, but Replit Assistant was the AI tool that worked inside the cloud ID and helped you do stuff. Then Replit Agent is the thing we've been talking about mostly. You prompt it, you keep prompting, you talk to it. Do those two things come together at some point in the future and become a single assistant that is more multimodal than what I currently have today? It feels a little arbitrary to have them split as they are. Again, might be silly, but I'm curious.

• 28:29Speaker 02

  No, absolutely. You're right. There are a few considerations here. The trade off on time and money, so and control. So we still have quite a bit of developers on the platform that feel like agent is taking quite a bit of control from them.

• 28:43Speaker 00

  I see.

• 28:43Speaker 02

  Whereas this citizen developer type, kind of like you, you know, is fine with that. Yeah, let it have control, let it make the decisions. But if if I'm a developer. At a company and I want to use a certain stack and all of that. I want more control and assistance also cheaper is five cent, you know, five cent a request.

• 29:04Speaker 00

  Versus twenty five cents, I believe, for agent.

• 29:06Speaker 02

  First twenty five cents. And also, some people go back and forth with that. For example, I want to tweak UI. I don't want to pay twenty five cent for tweaking UI and you can do it in system pretty quickly. I think we want to have like a little more variable pricing on agent, a little more variable time limit as well, like, you know, agent should know if I'm, if I'm doing a UI change, I shouldn't have to restart the server and doing all of that stuff. So I think it will converge, but it's going to be quite a bit of work. One thing that L L M's actually still lousy at is classification. I mean, the reason OpenAI has this dropdown with all these models and all these capabilities is because they can't really know which one you need given a query. I mean, that's the ideal situation, right? Sam talked about it on Twitter.

• 29:51Speaker 02

  It's like, we're going to put everything together. You don't have to kind of worry about all of this. But the reality is the L L M's are actually kind of bad at classifying these queries today.

• 30:01Speaker 00

  Talk to me a little bit about the future. I'm not going to ask you for a three year plan because God knows AI is going so fast, but quarters two and three this year. What are you excited about? What are you looking forward to? And what hints can you give us about where Replit might be heading?

• 30:14Speaker 02

  Agent V two, which is now out for our beta, what we call explorer users, is a lot more autonomous than Agent V one. It will do a lot more work per turn and people are a lot happier at it. The metrics we're seeing, like the deployment metrics are kind of insane. We'll announce them when we, when we launch the product. So we feel like we're headed towards more and more autonomy. Want to give, again, want to give customers a choice. Oh, I want to be sitting in front of my keyboard and do iteration, be really fast on that. Or I want to send this off to do it in the background. I want this change or when this app made, and I'm going to say, okay, yeah, I'm willing to pay this arbitrary number, you know, of, of whatever it takes and like go build this,

• 30:58Speaker 02

  test it yourself and then come back to me when you're done. And like, if you take fifteen minutes, thirty minutes, an hour, you know, I don't care. Just build this thing for me. So, you know, I sent this tweet earlier today. Hey, would you want Replit to be more autonomous? I always ask questions about our product. I've never seen such an emphatic yes, like people want more of that because they want to be able to send off requests on their phone and for it to like. For them to go shopping, come back and like their phone is already built. That the other thing I'm excited about is being able to unblock people like sometimes with all of all the products, the cursors, all of that out there.

• 31:39Speaker 02

  Sometimes you just get to a really funky place with vibe coding, what's called vibe coding, where you're not really looking at the code. You get to a place where actually you can't make a lot of progress. And so we were working through solutions to that. And you know, it might be kind of, um, you know, having, having like more compute, putting more compute in via reasoning models. Maybe even having some concept of a human in the loop. But I think that's a crucial thing to make sure that people never get really stuck and always can make progress.

• 32:10Speaker 00

  I mean, just to give an example of this, and by no means a diss, but I was having a problem with um, replit agent because it was saying. Do you see this? And I kept saying, no, I see a screenshot. It turns out I needed to open the web view tab.

• 32:22Speaker 02

  Yeah.

• 32:23Speaker 00

  So quite literally, I asked agent, do you want me to send you a screenshot? And it was like, I don't know. Okay. So I literally imported a screenshot. I'm like, I have no idea if this will work. And then it was like, oh, go to web view and then open that tab and then I sorted it out. But I mean, as a non-developer, I just don't have that built in sixth sense of knowing what I might be missing. And so I think unsticking people is great. So.

• 32:45Speaker 02

  Right.

• 32:45Speaker 00

  Sounds like better agents.

• 32:47Speaker 02

  Yeah.

• 32:47Speaker 00

  Do it on the go. And then also um, ways to help people that are less uh, developer inclined, not get stuck. That's a great mix.

• 32:55Speaker 02

  Absolutely. Yeah.

• 32:56Speaker 00

  And as the models get better, replit gets better, right?

• 32:58Speaker 02

  Yeah, I mean, that's, that's the exciting thing about, about the space. We're just riding this wave.

• 33:03Speaker 00

  Oh, that's so exciting. Thank you very much. Um, for folks who want to find replit, what's the URL? And what is one role you are hiring for that you're looking for candidates?

• 33:11Speaker 02

  Oh, awesome. R E P L I T dot com. And in terms of candidates, uh, really engineers, uh, AI engineers, platform engineers, finance engineers, so replit.com forward slash careers.

• 33:22Speaker 00

  Have you considered using replit for replit?

• 33:24Speaker 02

  We do a lot. I mean, um, one of the most exciting things is like, um, one of our HR uh, folks are building a lot of HR tools with a replit, which is wild.

• 33:33Speaker 00

  And people keep saying that SaaS is going to be just fine in the AI era. Well, we'll see. I'm John. Thank you so much for your time. I really appreciate it. Everybody. Talk to you soon.

• 33:42Jason Calacanis?

  All right. We all know if you're a founder or even if you're a small business, you're thinking about your company 24/7, 365 days a year. That's the life of a founder. This is not clock in, clock out, nine to five gig for you as the business owner. So when you're hiring, you want a partner that's as equally as committed as you are. And that's of course LinkedIn jobs. LinkedIn jobs is like your co-founder. They're going to make it so simple for you to post your jobs for free on LinkedIn, where there are 1 billion members. You're going to be able to share what you're posting and actually keep all the promising candidates organized in one place. And also LinkedIn is going to help you quickly write a job and get it in front of the right people,

• 34:26Jason Calacanis?

  whether you want to post for free or use some promotion to get it in front of even more qualified applicants. So do me a favor. Don't take my word for it. I mean, you should. I know what I'm talking about is where I find my great people. But just understand that 72% of small businesses using LinkedIn said that it helped them find the best candidates. So find out why more than 2.5 million small businesses already use LinkedIn for hiring. So here's your call to action. Post your job for free. Why wouldn't you do it? It's free, F R E E. That's a good price. LinkedIn.com slash T W I S T. Once again, that's LinkedIn.com slash T W I S T to post your job for free. Terms and conditions do apply.

• 35:12Speaker 00

  Hey, everybody. Welcome back to Twist. This is Alex and we have yet another Twist 500 interview today. And this is one that I'm incredibly excited about because we are once again going back to space. Yes, we're going to talk to a company called Starfish Space, which I think could have a very large role in making the in-orbit economy more economically viable and therefore more investable and therefore more active. You know that I'm a science fiction guy, so I'm always looking for companies that are going to take me eventually into orbit. Now, this company isn't that, but it is going to make the process of having people up in space just a little bit easier. So please welcome to the show. It's Austin Link from Starfish Space. Austin, hey, how you doing? I am doing well. I'm excited to be here with you, Alex. Hi, I'm so glad. So I want to talk about Starfish because of all the companies out there,

• 35:56Speaker 00

  you guys are doing the most like blue collar space thing of all time. You're essentially building, as far as I can tell, the in-orbit mechanics. So tell me a little bit about the company's vision and where you are today.

• 36:10Speaker 05

  Yeah, we are. We are kind of blue collar in a way. Our, um, one of our leads for the Otter program, a guy named Cappy, just his title on LinkedIn is Cosmic Garbage Man. Kind of what we do. Uh, but, but the reason why we do this is because I and my co-founder Trevor, just like you, just like people all around the world, are incredibly excited and passionate about humans going out into the universe. And to go out into the universe, we get to discover and explore and dream about what's out there. And I think sometimes we get to discover and explore a little bit about ourselves in the process. So as we do this,

• 36:51Speaker 05

  we've long been driven by large government organizations, by NASA paving the way and taking folks to the moon and landing rovers on Mars. But in the industry, we've seen, especially over the past couple of decades, that sometimes when large traditional organizations are, are trying to advance the boundaries of what's capable as we go out in the universe, it gets expensive. And it's, it's not always the most efficient development process. And so the reason why Trevor and I started Starfish Space is we said, hey, we have the same dream as so many people that like, let's go out in the universe, let's push the boundaries of what humans can do. And we think to make that possible, it has to be driven by the

• 37:34Speaker 05

  economics. Humans have to be getting more value from going into space than, than what it takes us to go into space. And that's why we've chosen to do Starfish Space as a business, because that really aligns our incentives as an organization with the overall problem. And it, it drives our approach as a business. So what we do as Starfish Space is we build, um, as our initial product, a satellite called the Otter, which is designed to grab and move other objects in space. It turns out that useful, and, and maybe we'll talk in a little bit about why that's useful. Um, but as we do that, it's not a unique thing to us. NASA sent up a space shuttle full of astronauts to do a mission

• 38:19Speaker 05

  like this, which you call satellite servicing, in the 1980s. So we've been doing it for 40 years. There's

• 38:25Speaker 00

  I bet that was cost effective, easy, and a very quick mission from NASA.

• 38:31Speaker 05

  You think you're saving a $500 million satellite. Boy, this is going to be a great deal, but when you have to send up seven astronauts in a space shuttle to do it, not money positive anymore.

• 38:40Speaker 00

  No, no, definitely net negative. Uh, but, but drill in a little bit on what Otter is going to do because it has kind of two functions in its current design, if you will.

• 38:50Speaker 05

  Two core missions that we go after with the Otter. One is satellite disposal. So you've got a satellite in low Earth orbit. It's died. It's just a cannonball threatening your other satellites now. We'll get it out of the way. Second core focus for us is what's called satellite life extension. You've got a valuable satellite. It's run out of propellant. It's drifting away from the orbit it needs to be in. We grab it. We bring it back to the orbit that it needs to be in.

• 39:17Speaker 00

  But to be clear, to, to de-orbit a satellite, to push it into an orbit where it eventually decay and then just break up and kind of atomize itself. What you do is you apply pressure and pushing it down. And when you're trying to increase the life of these satellites, my understanding is that the Otter becomes effectively something that latches on and then gently pushes. So you're either pushing up or pushing down. So you've created like the world's most advanced staircase.

• 39:43Speaker 05

  Well, and what gets tricky is some of the orbit dynamics is part of this. And, uh, this video is, is going to humorously not follow some of the orbital dynamics. Um, but when you want to go down in space, you don't actually push down. You push backwards and that drains energy from your orbit and it goes down.

• 40:02Speaker 00

  So essentially you just remove energy and then, okay. Interesting. So, but what, what, what would happen if you pushed down though? Now I'm curious.

• 40:10Speaker 05

  Um, you just kind of wobble around your orbit and you don't even notice it.

• 40:13Speaker 00

  Oh, well, this is why we have experts on the show, ladies and gentlemen. So you push back at a satellite or now I presume instead of up, you're pushing forward.

• 40:22Speaker 05

  Yes.

• 40:23Speaker 00

  Okay. And there are fancy terms for this, but the gist, Austin, is that there's so many satellites in orbit today and the pace at which we're adding satellites seems to be at an all time high. So I presume there's more of these dead cannonballs than ever. And also, there's probably more people who want to keep their existing satellites serviced and in orbit longer.

• 40:40Speaker 05

  We're only putting up more and more in space. And that's both because it's becoming more affordable and you see trends like smaller satellites and launch making that happen. And also because of all the value it's providing to folks, whether it's earth observation, whether it's communication, really tremendous things going on in space. But as we put more up there, things that were previously not a problem because space is really big, become problems. And also, we want to have more capabilities. If we can interact and recycle and assemble satellites up there, that's more and more that we can do in space and build a space economy.

• 41:17Speaker 00

  Absolutely. So, one thing I've been impressed by, and one reason why I wanted to have your company on our Little Fun Twist 500 list, and I wanted to have you on the show, is because it seems like your pace has been really quick. Company founded 2019, and in 2023, just four years later, more or less, you guys actually launched a, what I think is a proto version of the Otter called the Otter Pup, which is, I believe, the technical term for an otter baby, which is adorable. I have young kids, so I've read a lot of books like, you know, giraffe, cat, cow, yeah, so. But that mission, when your Otter Pup made it up into space, didn't go as planned. So can you tell us a little bit about what happened and then the recovery process? Because I think it's an illustrative story for how hard what you're doing is.

• 42:01Speaker 05

  Yeah. It's hard to start a company. And something that was totally foreign to me in starting Starfish Space and totally foreign to my co-founder. And you put so much of your hopes and dreams into a project and so much of your sweat equity. And then, things happen. Things always happen. And so for us, we spent, we raised our seed round, we spent a couple of years designing and building and testing the Otter Pup, a small satellite designed to dock with another satellite orbit. And this is going to prove out all of our key technology and our software. And we are sitting on top of another satellite that is attached to SpaceX's Falcon 9 rocket.

• 42:44Speaker 05

  And SpaceX does their job often awesome and they pop us off into orbit. And we're sitting on top of this other satellite. And they had an issue and they had an issue that led to. Their, one of their thrusters being stuck open and they drained all of their onboard propellant. Within a few hours after we separated from the rocket. And spun our combined satellite stack up and to their credit. They realized this was going on. They took some emergency actions to save us, to give us a shot and popped our satellite off. But that means that our company began our life as a spacefaring company. With our satellite spinning at 330 degrees per second.

• 43:29Speaker 00

  And for folks who don't know, there's only 360 degrees in a circle. So you're spinning around almost every second.

• 43:34Speaker 05

  We're spinning around almost every second. Our satellite is designed to spin up to two degrees per second. So it's 150 times faster than we were supposed to spin.

• 43:43Speaker 00

  Well, you wanted to stress test. I mean. Stopping you here Austin to ask a puckish question, but when you're doing a demo and a test like Otter Pop. What percentage of the mission is you guys. Learning as a company by putting something up in orbit to actually test? And what portion of the effort is to show other people that you've built something that works? I've always been kind of curious about these early missions and the proof outs and who they're really for.

• 44:08Speaker 05

  Um, for us, we do it all for our own test and development. And when you're like, we sell to. Oftentimes other engineers making decisions. And so. For us, the, the marketing or the selling is almost just showing that we've done a good engineering development process and, and, and builds their confidence and ups.

• 44:29Speaker 00

  So, okay. So you're spinning 30 degrees a second. You guys actually managed to slow it down. Did you slow it down all the way to, to stationary in terms of spin?

• 44:38Speaker 05

  Yeah, so it was a it was a month long journey to get in contact with the satellite, to get it power positive by deploying the the solar panels, to get it thermally stable, and to figure out, okay, can we figure up, can we can we stabilize the satellite? Can we detumble it? And as you're going through this process as a founder, there's a part of you as a founder that's like, I don't think this is good for the business right now. And there's a part of you that is sort of the the program manager that that is saying like, oh, my gosh, we put so much effort in, this was such a dream mission, and now it's now it's at risk and it might not happen. And then there's a part of you that is the engineer. And the party that's the engineer is like,

• 45:22Speaker 05

  oh, man, we might have to, we might try to save this. Like, this is going to be exciting. I can't wait to tackle this problem. And so in a weird way, you're you're very stressed out and you're very worried, but you're also really energized for the process. And we eventually adapted some of our simulation and our algorithms to, instead of going to dock a satellite, to use magnets onboard the satellite and gently push off of Earth's magnetic field, and over time stabilized the satellite so that it wasn't rotating anymore.

• 45:53Speaker 00

  That sounds, from my uneducated perspective, like a very slow process.

• 45:58Speaker 05

  Um, it is, it is kind of slow, but it it did end up playing out over just like a week once we nailed the algorithm.

• 46:04Speaker 00

  Okay, so a week's a week is shorter than I expected, but I still wouldn't call a week quick. Is a week quick in satellite terms?

• 46:14Speaker 05

  Takes you over a year to build these things, so a week is a week isn't too bad.

• 46:18Speaker 00

  Got it. Okay, what this is again, slightly puckish, but like, how much of the company's capital was tied up in the Otter Pup mission? So when it was doing the wee, how worried were you about accidentally, I mean, kind of crashing the business just because you had a bad luck on your space tug?

• 46:34Speaker 05

  You always worry about that as a founder. You're always worried about how is our business going to survive this thing that's going wrong? And what can kill our business? We had, as we'd made progress with the Otter Pup and as we'd made progress with customers, we'd gone out to fundraise prior to the launch of the mission. And so for us, fortunately, it wasn't like, man, we have to get something or else we're all out of our jobs. It was in fact, as soon as it started going wrong, like, all right, well, let's, let's talk about what path the business has forward to make sure that we can still succeed in spite of a mission going a little off the rails.

• 47:07Speaker 00

  And are you referring to your Series A1 round? That was about 29 million?

• 47:12Speaker 05

  So that particular round that came just before the Otter Pup mission was our, our Series A, the $14 million fundraiser.

• 47:21Speaker 00

  Okay. And so the A1 came after that. Got it.

• 47:24Speaker 05

  Yes. Yeah. So we ended up raising about $29 million late last year.

• 47:29Speaker 00

  Now, 2026, you guys are doing your first official mission with Otter. Talk to me about the partner for that and what the goal is.

• 47:37Speaker 05

  Yeah. So as we sort of pulled rabbits out of the hat on the Otter Pup mission, it turns out it built a lot of confidence in our customers that they didn't want to go, hey, this thing is going perfectly. They kind of assumed it's your first mission, it's risky, it's not going to happen. They just wanted to see us solve some problems and do a good job in the engineering. And it led to us signing three major contracts over the last summer to launch Otters in 2026. And so the first of those is with a commercial company called Intelsat, which is one of the largest geostationary satellite operators in the world. The second is with the US Space Force and the third is with NASA. And so we've got three Otters launching in 2026.

• 48:23Speaker 05

  And that is dominating a lot of my efforts right now. After this, I got to go downstairs and say, all right, how's the assembly going? How's the test going? What can I do to help?

• 48:32Speaker 00

  So paint me a picture of your factory floors, Will. I know you're up in the Seattle Redmond agglomeration. Are there three Otter pups like in various stages, sorry, three Otters in various stages of completeness downstairs that you can kind of go see as they're sequentially built?

• 48:48Speaker 05

  So we actually work heavily with a series of manufacturers. As we build the Otter, there are certain things that are really unique to our satellite that we build in-house. So there's some specific hardware that you go downstairs and you'll see. But you won't see anything that looks like a half-assembled satellite. Because for many pieces of our satellite, we work with external suppliers. As an example, on our OtterPup mission, OtterPup never existed at our facilities. It existed at our partner's facilities, a partner called AstroDigital, who is really great to work with on that mission. And we build our hardware, we build our software, we go down and we install it in their facilities. Allows us to move a little faster.

• 49:30Speaker 00

  Over time, though, do you think that off-the-shelf parts and their construction will move in-house? Or will you guys stay, I don't know what the space term here is, but we would call it asset-light in some ways, perhaps in a different context?

• 49:47Speaker 05

  In general, as we do things that are really unique to us, it makes sense for those to be in-house. As we do things that are really similar to what other folks do and buy, then if we work with an outside supplier, we can use their whole supply chain, we can use their reliability. We don't have to upfront all of the non-recurring engineering costs. So we do see value over the long term of working with folks externally.

• 50:11Speaker 00

  That actually seems very bullish to me, that there is a sufficiently robust value chain in satellite construction that a company like Starfish can simply focus on differentiators, other technology. But then you can kind of like build that on top of what already exists. So you can go from incorporation to space in four years versus an old timeline, I don't know, 40 or whatever it would have been. But it feels, it feels kind of awesome that you don't have to build the whole thing in-house, even if it does make for a less good, like, you know, camera shot from the staircase.

• 50:42Speaker 05

  It is awesome. It's an awesome set of infrastructure from the satellite manufacturing to folks building radios or cameras or thrusters, even to SpaceX and a launch vehicle.

• 50:52Speaker 00

  Yeah.

• 50:52Speaker 05

  And hopefully it all grows with us as we continue to do more with Otters.

• 50:57Speaker 00

  So I'm curious. About the economics of this because once you have a satellite in space, every otter will have its own life expectancy. It's gonna be scooting around doing things. Um, are you guys gonna own them as a company? Or are you gonna sell them to say a NASA or another company that then will own it themselves? And then if they own it, do you operate it for them? I was trying to figure out where the money flows.

• 51:19Speaker 05

  Yeah, so we own it and we provide services to customers. Um, and as an example, Northrop Grumman is a company that began providing the first life extension mission in 2019. They're getting paid $65 million to extend the life of a single satellite for five years. If you're, if you're building a large geostationary satellite, you can be making 30, 40, 50 million dollars a year.

• 51:44Speaker 00

  Got it.

• 51:44Speaker 05

  An extra five years of lifetime is huge. Uh, these are, yeah. And so if we provide those services and we do so at a cost point that is less than that, we can build a good and sustainable business.

• 51:56Speaker 00

  You know, Amazon, uh, Jeff Bezos always said that your margin is my opportunity. My thought is even with space economics where they are just today with launch costs as they are and so forth, $65 million for a single life extension mission seems like a price point that you can beat pretty comfortably and still make a lot of money.

• 52:16Speaker 05

  You can with the right approach. You have to do it with a small satellite. And that's where a lot of our unique approach at Starfish comes in and our unique technology comes in. Getting two satellites that are moving full, full speed to gently come together and grab onto each other is really, really tricky. Folks traditionally solve that with a bunch of hardware, but hardware is mass, is volume, and it makes a big expensive satellite. We focus on solving the problems with software that can dock much simpler satellites to be able to, to do these missions at a price point that closes the business case.

• 52:53Speaker 00

  So this is kind of the three points to the the Otter Project. Cephalopod, which is your autonomous guidance and control software. Then the Nautilus, which looks like a big suction cup, as far as I can tell. And then I actually don't know how to pronounce the other word. Setasian?

• 53:10Speaker 05

  Setasian? Cetacean, which you wouldn't know unless you've been thinking about whales a lot recently.

• 53:16Speaker 00

  I've read that word hundreds of times. I never had to say it out loud until right now. And I looked at all my notes and I went, crap. I don't know how to say that. Anyways, over having to do with whales, over having to do with squids, and then Nautilus, of course, is a famous. Yeah. So those are the three components that come together to make this happen. How hard was it to write code to get these two satellites moving, I don't know, three to eight kilometers a second up there? To come together?

• 53:44Speaker 05

  It's incredibly hard. And maybe just to frame what we're trying to do, we're trying to take a satellite. Even the Otter Puff is the size of a microwave. And we're trying to steer it into dock using something called electric propulsion. And electric propulsion gets great gas mileage. But the force that it puts on your satellite is the same force as like a housefly sitting on your hand.

• 54:08Speaker 00

  Oh, it's even less than I thought.

• 54:09Speaker 05

  And so we are steering our satellite through space to go dock with another satellite by like strategically putting a housefly on one side or the other at different times.

• 54:20Speaker 00

  Is that is that effective because drag is so low in orbit? Because if you think about this, like that wouldn't work in a higher drag environment.

• 54:31Speaker 05

  Yeah, so we do have to do some things to account for drag even in space. Because the forces are so low and you have to account for the pressure that the sun puts on you just from its like photons and other particles. So what?

• 54:46Speaker 00

  The solar wind.

• 54:47Speaker 05

  Yes, yeah, so we have to account for all of that. But and then you have to predict the orbit dynamics like very precisely, very well in advance. But if you do that, then you don't have to have 24 different super powerful thrusters hanging off of your satellite.

• 55:04Speaker 00

  Right.

• 55:05Speaker 05

  And your satellite can be a lot smaller and you can close the business case.

• 55:09Speaker 00

  So why isn't electric propulsion better by now? Is that just limited by physics? Cause I feel like I've heard about this propulsion method for some time. And the fact that it's now as powerful as a housefly landing on me. Well, it doesn't impress. I feel unimpressed by that. And maybe that's my personal ignorance shining through. But when is it going to be the size of, I don't know, a ladybug's weight on me or something?

• 55:33Speaker 05

  It all, it's all a power limitation because to do this electric propulsion, you have to rip electrons off of xenon atoms. And you have to accelerate them through a powerful electric field. And they have to get going really fast because the faster you throw stuff out the back, the better gas mileage you get. If you use fuel like a rocket, there's all kinds of power and energy just in that fuel. But with electric propulsion, we've got to get it all from the solar panels. And you end up just really power limited from your getting power from your solar panels and your battery. And then you have to rip off these electrons and launch them out the back. And, and that forces, I mean, we use dramatically lower thrust than, than what say the state

• 56:16Speaker 05

  of the art in electric propulsion is just because we have only a certain size solar panel.

• 56:21Speaker 00

  So essentially it's photons in, electrons out.

• 56:24Speaker 05

  Yeah.

• 56:25Speaker 00

  Okay. That's actually a pretty.

• 56:27Speaker 05

  Full, full xenon atoms out, but ionized xenon atoms.

• 56:30Speaker 00

  Oh, so you strip the electron off.

• 56:33Speaker 05

  And then you have to launch both out your back so that your spacecraft doesn't build up a wonky charge.

• 56:38Speaker 00

  Oh, and oh, of course, right. You can't hold on to all the spare electrons because then you'd have like the equivalent of like the worst static electricity in space.

• 56:46Speaker 05

  Yep.

• 56:47Speaker 00

  So what I love about this is I did not know that the solar wind actually is strong enough to impact orbits. But I think this goes to show that the space is, is more complicated than enterprise SaaS. But. I mean, it seems doable. People are doing it. Tell me a little bit about how optimistic you are about startups in the modern space economy and how fast things are progressing. Because to me, again, uneducated on this, it seems to be going pretty quick and I'm very impressed and excited, but I'm, I'm hoping that that's accurate.

• 57:19Speaker 05

  I think that there are a lot of startups doing amazing things in the space industry right now and SpaceX leads the way and what they're doing is really incredible. But you see folks like Rocket Lab, folks like Planet. Firefly fricking landed on the moon recently.

• 57:35Speaker 00

  I know.

• 57:36Speaker 05

  That's amazing.

• 57:37Speaker 00

  How did the second time a private company landed on the moon not become front page news? It's almost like there's something else going on in America right now that could be consuming a modicum of oxygen. But I saw that and I was blown away. I mean, I grew up on, on like, I grew up watching Apollo 11, the movie, you know, like, I mean, that accomplishment was a moment of like, peak national pride. And now some, you know, some for-profit nerds, said with love, are like, we're gonna do it too. That's awesome.

• 58:09Speaker 05

  It's, it's incredible. And, and I think it's all bringing to bear some of what we hope to bring to bear, which is, well, let's build an economy that gives us a reason to keep going back out into space.

• 58:21Speaker 00

  And that brings me to my only kind of quasi awkward question for you, which is, how do you guys stack up against OrbitFad? We've talked to them, they're cool, they want to do in-orbit refueling. And one thing that I'm confused by is just what's the better method? Bringing propellant to satellites and then trying to give them more so they can push themselves? Or have an otter come by and boop you with its nose and then get you back up?

• 58:44Speaker 05

  So the OrbitFad folks are actually pretty good friends of ours. I know Dan pretty well. We, we share at least one investor, maybe a couple of investors. And in many ways, a lot of what we're doing is complementary. That refueling satellites can be great and the OrbitFad folks are building these refueling ports, but they don't exist on satellites today. And so we focus on, all right, we have to go provide services to satellites today. And someday I think satellites will have refueling ports and they will have ports for you to transmit data and power back and forth and the solar panels will be easy to replace and you'll just strap on the latest compute capability to add more to the satellite.

• 59:28Speaker 05

  Like someday, that's how satellites are going to work. The otter is a big part of making that possible.

• 59:33Speaker 00

  Well, I love that you guys are in competition because I had a great chat with them and I've really enjoyed you. And so I'm kind of hoping that like, I don't know, space is big. Maybe we can all be friends and get along and go faster together. Before we go, where can people find the company online? And what is a role you are currently hiring for that you'd love to get the word out about?

• 59:50Speaker 05

  Yeah, so starfishbase.com. Come find us. We're on X, we're on LinkedIn. The big hiring push right now is on the software engineering side. You have to write a lot of software to fly satellites. And you don't have to have written a lot of satellite software or a lot of rocket software. You can be writing B to B SaaS software and you can come in and be an awesome contributor to a satellite mission. And it's, it's not always the perfect opportunity. It's, it's probably a little longer work hours. It's probably a little bit less pay. We, we try to do our best on those. But it's a mission that is really exciting to be a part of. Maybe sometimes more exciting than building a CRM tool.

• 1:00:32Speaker 00

  Yeah, I was gonna say, there's that old Steve Jobs quote, you know, do you want to sell sugar water or come change the world? Well, do you want to add a feature to Slack that users will hate? Or do you want to send cool things to space? So if that's you, talk to the guys over at Starfish. Austin, thank you so much. And when the next launch goes up, please come back on the show, tell us all about it, what you've learned and let's get those launch costs down as low as we can.

• 1:00:54Speaker 05

  I'm on board with all of that. Thanks for having me, Alex. What a, what a pleasure.

• 1:00:58Speaker 00

  A real treat. Wasn't that fun, everybody? I freaking love talking to founders. After all my years of doing this, it has never, ever gotten boring because there's something special about someone who's trying to build the future that gives you just that dose of optimism that you absolutely need no matter what day of the week it is. And speaking of that, don't forget, Twist comes out on Mondays and Wednesdays. Days and Fridays, we go live. You can watch us create the show in real time. We even published the docket that I put together, thisweekinstarbucks.com/docket, so you can see us build the show and you can even follow along with our notes as we tape. My name is Alex, Alex Overon X. I'll see you tomorrow. Jason and I are going to be back to go over all the news. I'll see you then. Bye.