In short
Two-part episode on enterprise AI “owning your intelligence” plus autonomous drone delivery logistics. First guest: Josh Sirota (CEO, Aragon) argues companies should train and continuously update models on their proprietary data so workflows run faster and cheaper without stuffing every task’s context into frontier model prompts. He describes “company world models” using reinforcement learning (RL) and “post-training” to update model weights overnight via continuous learning.
Key claims
Aragon reduces token usage by ~100x and speeds workflows ~10x versus frontier approaches; it can embed company knowledge in weights so it doesn’t need to re-read email/Slack each time.
Notable examples
email triage and Slack-to-candidate dashboards; sending the correct “Josh” for an MSA by disambiguating identities; paying a Milton office-cleaning invoice via connected financial infrastructure (QuickBooks/PayPal/Venmo) and enterprise agent bank accounts. Pricing: $5 per million blended tokens; positioned as 1/3–1/2 the cost of frontier.
Guests
Josh Sirota (Aragon). Second guest: Etienne Louvet (founder/CEO, Iona) builds autonomous delivery drone systems for logistics operators, using BVLOS/Part 108 to enable remote supervision and scaling. Examples: white-labeled drone delivery for CMA CGM and Welsh Group; manufacturing/iteration in Galway, Ireland; aims to connect remote communities.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Chapters
Tap a time to open that second in VOIntroduction to Josh Sirota and Aragon
1:18 to 3:52
Discussion on Josh Sirota's background and what Aragon does.
“We also put half the show into the This Week in Startups feed just to tease you a little bit, give you a taste.”
Building a Company World Model
3:52 to 7:30
Josh discusses the concept of a company world model and its significance.
“But when we get to off-duty, two things.”
Understanding Continuous Learning and Model Upgrades
7:30 to 10:50
Exploration of continuous learning and how models can update efficiently.
“and therefore like our whole slogan is actually own your intelligence, right?”
Demo of Aragon's Capabilities
11:20 to 14:01
Josh demonstrates Aragon's capabilities in handling workflows and data integration.
“So I think first thing that we can go through is you guys had just brought up, you know, not checking email.”
Understanding AI Model Integration
14:01 to 15:12
Learn how AI models can streamline communication and task management.
“We're doing a really exciting partnership with a very exciting ERP company.”
The Role of Financial Services in AI
15:12 to 16:18
Explore how AI agents can interact with financial services for efficiency.
“So one is like, okay, I just went through probably, in this case, like, I don't know, maybe 50 emails or so.”
Partnership Ecosystems in AI Development
16:18 to 19:36
Discover the importance of partnerships in enhancing AI functionalities.
“And so then the thing that we are working on with like our partners is also building out this ecosystem inside of Aragon.”
The Importance of Data Weights in AI
19:36 to 20:25
Understand how data weights enhance AI performance and reduce costs.
“And so like, we just did this pretty fast, I would argue, like, it would be, you know, an order of magnitude faster than anywhere else.”
The Importance of Data Weights in AI
21:15 to 21:32
Understand how data weights enhance AI performance and reduce costs.
“And because it has in the weights, because it's updated itself, it doesn't need to go for that basically workflow and read all of my email, read all of the Slack.”
Owning Your AI Intelligence
21:32 to 25:29
Uncover the significance of owning proprietary AI data and models.
“And it should be incorporated into its logic at a more base level.”
Show all 37 chapters
Cost-Saving Strategies in AI Integration
25:29 to 28:00
Learn about the cost efficiency of integrating AI models in businesses.
“And it's always a great tell for a founder, for me, Lon, when I'm working with a founder, how crisply they can explain the business model.”
Streamlining Quote Requests with AI
28:00 to 29:12
Learn how Aragon automates the quote request process for startups.
“You can actually go and basically request a quote.”
Josh's Journey: From Rejection to Success
30:11 to 31:26
Hear the inspiring story of Josh's startup journey and the importance of mentorship.
“You want to do that, but it's going to be done faster, better, and cheaper.”
Using AI to Enhance Decision-Making
31:26 to 34:04
Explore how AI systems can improve decision-making in startup applications.
“And it's just great to be in business with you.”
The Future of AI and Human Collaboration
34:04 to 35:01
Understand the balance between AI capabilities and human oversight in startups.
“Let me know if you're ever in Austin and we'll go grab some barbecue with lawn.”
AI Advancements and Recursive Learning
35:01 to 35:50
Delve into how recursive learning in AI can lead to exponential growth in capabilities.
“We're going to have our agent throw that up on the website right away.”
Latest Trends in AI Development
35:50 to 39:28
Get insights on the latest AI models and their implications for various industries.
“We've been trying to have the agents run in a sandbox and in Amazon's rack somewhere that they need.”
Iona's Vision for Autonomous Logistics
40:01 to 42:00
Discover how Iona aims to revolutionize logistics with drone technology.
“So do you want to demo and show us what you built?”
Building the Autonomy Stack for Logistics
42:00 to 43:16
Learn how a new autonomy stack is designed to support logistics operators amidst competition from big tech.
“So this is, you're talking about the FedEx, UPSs, DHLs of the world.”
Understanding Part 108: A Game Changer
43:17 to 44:29
Discover what Part 108 means for drone operations and how it simplifies regulations.
“We can get, we can double click on, you know, these kind of things here on the show.”
The Shift from Visual Line of Sight to BVLOS
44:30 to 46:26
Explore the transition from visual line of sight flying to beyond visual line of sight operations.
“So it reduces the threshold and it creates a framework that is much more adept to a delivery with a drone, for instance.”
Creating a New Air Traffic Management System
46:27 to 48:26
Learn about proposed systems for integrating drones into shared airspace with traditional aircraft.
“They could be getting paid 30 bucks an hour.”
Drones in Galway: Connecting Remote Communities
48:27 to 51:15
Understand how drone technology is being utilized to connect remote areas in Ireland.
“I haven't been there in 25 years, but I drove from Dublin to Galway.”
The Market for Drone Deliveries: Light Cargo
51:16 to 52:49
Discover the focus on light cargo delivery and what items are most in demand from consumers.
“Like when you talk to your consumers, you know, you've got people in Galway, they're on farms, they're in rural areas.”
Drone Specifications and Capabilities
52:50 to 53:48
Learn about the specifications of drones, including payload and range capabilities.
“So that's significantly larger than the other folks out there.”
Innovations in Drone Charging Infrastructure
53:49 to 56:00
Explore the future of drone charging systems and the potential need for common infrastructure.
“So for instance, we also have demonstrated a prototype for loading and unloading the parcels autonomously.”
Innovations in Aerial Delivery
56:00 to 57:52
Discussing the unique value proposition of various aerial delivery models.
“So it takes off vertically and it transitions to horizontal flight so that we can have the convenience of the vertical takeoff.”
Safety First in Delivery
57:52 to 58:09
Emphasizing the importance of safety in aerial delivery operations.
“Listen, Ethan, continued success and be safe out there, right?”
Breakthrough Prize Experience
58:09 to 59:16
Sharing experiences at the Breakthrough Prize event and celebrity interactions.
“One of the first Burning Mans I'm at, there's like a bunch of signs.”
Conversations with Filmmakers
59:16 to 1:01:06
Engaging discussions with filmmakers about past experiences and future projects.
“Julia Chastain was at the table next to me.”
Ikiru: A Cinematic Masterpiece
1:01:06 to 1:02:39
Analyzing the themes and significance of the film Ikiru.
“People always focus on Yojimbo, Sanjuro, Seven Samurai.”
The Future of AI in Filmmaking
1:02:39 to 1:07:46
Exploring how AI is transforming the filmmaking process and its implications.
“to do one thing in his life that just counts.”
Nostalgia and Cultural Reflections
1:07:46 to 1:10:06
Reflecting on nostalgia through cultural references in media and personal anecdotes.
“to do certain functions and speed things up and make things cheaper to allow the creative people to do more with the creative side.”
Nostalgic New York City Venues
1:10:06 to 1:11:28
Explore iconic NYC locations from the early 90s and their cultural significance.
“And then there's a bar, Walker's, Bubby's, the Odeon, Tompkins Square Park, the Guggenheim.”
Overview of Dust Bunny Movie
1:11:29 to 1:12:33
Discover the intriguing premise of the film Dust Bunny and its themes.
“It's Brian Fuller, who people remember he did that show Pushing Daisies.”
Healthy Eating Insights: Tinned Fish
1:12:34 to 1:14:46
Learn about the nutritional benefits of tinned fish and personal dietary choices.
“I mean, if they are mature 10 year olds, it's a little bit of violence.”
Exploring Anchovy Fasting and Recipes
1:15:17 to 1:16:53
Discussing anchovy fasting trends and sharing recipes for healthy eating.
“And there is something about a shape I've been since my 20s.”
Transcript
Automatic transcript. May contain errors.0:00Hey, can we just have a better CEO than Jason? Can we have a better board of directors than your board of directors because they have all information? You're saying now you're building that, Josh. Yeah, exactly. What we're calling this is a company world model to make that vision true, an algorithm that can actually have the right reward functions. So we can get up to 20 parcels in the drone. 40 plus pounds. Holy cow. If you're already doing logistics, you could just add drones to your fleet. This is kind of like going from level two autonomy to level four. Yes, exactly. It's happening, folks.
0:36This Week in Startups is brought to you by Pilot. Focus on your product. Let Pilot handle your bookkeeping. Pilot provides the most reliable accounting, CFO, and tax services for startups and small businesses. Head to pilot.com slash twist and get$1 ,200 off your first year. Vanta. Compliance and security shouldn't be a deal breaker for startups to win new business. Vanta makes it easy for companies to get a SOC 2 report fast. Get$1 ,000 off for a limited time at vanta.com slash twist. And Quo. Quo, formerly Open Phone, gives you a clean, modern way to handle every customer call, text, and thread all in one place.
1:15Try it free at quo.com slash twist. Hey, everybody. It is 420. Lon Harris is here. He's ready to go. Yeah. And it's Twist. and we got a full docket. Monday, Wednesday, Friday. I do Monday and Friday. I do some Wednesdays. You're here on Wednesdays. And the This Week in Empire is growing. This Week in AI drops on Wednesdays. We also put half the show into the This Week in Startups feed just to tease you a little bit, give you a taste. You get your first 30, 40 minutes. And then we remind you, hey, go over to the This Week in AI show and subscribe there. Like a dealer. The first half is free and that's how we get you.
1:54Sure, sure. A 420 perfect analogy. And we got a lot of important things going on. There's a lot of notes you're going to need to take. This week in startups, our core pillars are tactical and practical and experts only. Those are the first two pillars. And when you have tactical, practical information and experts only on the show, man, this means you're going to get a lot out of it as a founder. So I want to just take a minute for plod. I'm wearing it on my wrist today. I've been enjoying my wrist. And I just press this little button, Lon. And, oh, the red light's on. Yeah. And now, if I think of something, oh, the next This Week in show that we're going to launch, soft launch in a couple of weeks, make sure that blah, blah, blah, blah, blah.
2:39Oh, we're not giving it away. And that we blah, blah, blah, blah, blah, which is what I sent in text today. But I will do this frequently to myself. And I did it when I was on the road. I was on my way to the Breakthrough Prize. I'll talk about that in Off Duty. We always do Off Duty at the end of the show. and I'm wearing this thing and just I'm going through airports. I'm on a plane. I'm getting my Starbucks. Bing. Da-da-da-da-da. I get my go-to. I get my action items. Now, when it goes into the tray and I charge it, which you only need to charge this thing. What are you charging? Like every 10 days or something?
3:09Yeah, that's one of those standout things. Once a week, I remember to drop it on that charger and that's all. Otherwise, it stays just on my suit jacket right here. And if you want to get applaud, we have a deal for you. Yeah, if your work relies on conversations, We want you to get a Plod Note Pin S. You can check it out yourself at plod.ai slash twist and just use the code twist and you'll get 10 % off on your Plod Pin. Let's get to work. Let's get our first guest on. We'll do off duty. I'll talk a little bit about the venture training program I've launched in Austin. And I'm going to talk about the venture training program.
3:45We got so much news today, too. Palantir had their big statement. Moonshot launched Kimmy K 2.6. So much to talk about. There's a lot to talk about. But when we get to off-duty, two things. I'm going to talk about the Breakthrough Prize, the Science Prize with unlimited celebrities, the most powerful investors, founders in the world, all in one room. Somehow I made it in. I'll explain that. Three or four besties were there. And I'll also talk about our venture training program. I've now made it six seats because there's so much demand. We have a couple of hundred people applied already. So I'm going to get 600 applicants.
4:19I'm going to pick the top 1%, and we'll talk about that at the end of the show. But let's get to our first topic. Yes. And or guest. It's our guest. We're bringing on Josh Sirota. He's the CEO of Aragon, Jason. I know him. They're sort of an enterprise version of OpenClaw, maybe an AI operating system for enterprises. Their big concept is they want to help companies own their intelligence. So they're connecting a company's existing tools and data to custom models that have been trained on their data and purpose-built autonomous agents. So Josh, thanks for being here. Yeah, great to see you. So I always like for the founder to tell us why they're creating this, how much it costs, who are the customers, and do a demo.
5:01So let's get to work. Why did you build it? Who are the customers? What does it cost? And let's demo it. Demo or die here on Twitch. Okay, amazing. Let's do it. So as far as why we built Aragon, Basically, like our view on the, you know, on what's happening right now is as follows. So you are taking a, you know, frontier model. And as we saw with, you know, OpenClaw or, you know, Hermes or, you know, different other innovations, there's been a lot of research and a lot of work that has went across these harnesses around these frontier models. And basically, for us to be able to execute, you know, all of these tasks inside of companies, what we're doing is collectively creating these memory systems with these like.mds and trying to basically pull all of this into the context of a model.
5:56And so what you see is if you look at the number of tokens in the world, it's like a thousand X input tokens versus output tokens. And lots of workflows, like I don't know if you guys saw Uber CTO just said that they blew past their whole budget, you know, already in April. So it turns out that even with all of these innovations and, you know, harnesses, it's actually like quite expensive, depending on the workflows that you want to do, to try to pull into context, you know, everything that you need into the model. And so what our view is and like what we're researching is basically an RL algorithm that can actually take a lot of the things about your company and have those things be in the actual weights of a model that you own that's specific to you and your company that can update itself overnight.
6:51and therefore it actually begs the question of like okay you know do you what do you need to use the frontier kind of closed source models for versus like what can you actually own on top of a lot of your proprietary data sets and so that's kind of why we started this which is we just think there's a huge amount of like inefficiency as far as how we're executing tasks and we don't necessarily think you have to use these frontier kind of closed source models right for everything and what we do is basically like have this RL piece that can actually take certain parts of your company and put it into the weights of a model that you own and therefore like our whole slogan is actually own your intelligence, right?
7:33This should be your own. I think that this is exactly where things are going. I recently tweeted like, hey, we're on a bit of a collision course and we were talking about SLMs just last week with Rob. small language models can be produced and run on local hardware. And if they have all your data in it, well, what becomes the difference between an open claw and all these skills being made that then go have a context window, have memory, or it all being dumped directly into an SML you own? So you take your entire sales database, you take your entire Slack, you take every email, and you put it into this sml and this is the the vision of ultron we had 80 days ago we're sitting here in ao 84 i think yeah lon and we were like hey can we just have a better ceo than jason can we have a better board of directors than your board of directors because they have all information and you're saying now you're building that josh yeah exactly so i think like a couple things to add there is like what we're calling this is a company world model.
8:46Most of my team is, you know, ex-DeepMind and, you know, PhDs from Berkeley Air Research Lab. And a big thing on the frontier side in research is really RL, right? Reinforcement learning. And what doesn't exist yet is like to make that vision true is an algorithm that can actually have the right reward functions and be able to update the weights of the model in a frequency that actually makes sense for a business, right? So instead of updating this model once every three months and it costs a lot of money, can this model actually update itself overnight based on the interactions that it has with people?
9:28Where do you run the model? What models are you building this on? Kimmy, DeepSeek? What do you use as your foundational model to then merge? What's the term when you put the data in for my company every night? What do you call that? Is it a build? Is it post-training, pre-training? How would you explain this? Details matter. Building a great company is all about paying attention to details, and they add up along the way. That's why today's episode is brought to you by Quo, Q-U-O, the smarter way to run your business communications. And I use this product. We love it at launch. Why? Well, I don't want my sales team using their personal phones.
10:08I want them to have their own phone number. For some phone numbers, we want there to be one phone number that we can put at the bottom of an email on the bottom of our website so that no calls get missed. And it's so much easier to service an existing customer than it is to find a new one. And Quo is the way to do that. They're being used by over 90 ,000 businesses. Everything from solo operators to at-scale teams love Quo. And it works wherever you are. Android, iOS, desktop. It's going to work everywhere. So sign up for Quo's smart system and make sure no more opportunities slip through your fingers.
10:40Try Quo for free and get 20 % off your first six months when you go to Quo.com slash twist. That's Q-U-O.com slash twist. Quo. No missed calls, no-miss customers. So this is all called like post-training. It all falls into the category of post-training. And then the method is called RL, reinforcement learning. And as far as the model that we've found that has like, you know, very good kind of base is Kimi. So we're also very excited for all of the announcements that they're going to be making. And as far as the concept, right? It actually doesn't yet exist to do the updating. This is what's called continuous learning.
11:21So demo or die time, Josh, let's do it. Okay, let's do it, guys. All right. So I think first thing that we can go through is you guys had just brought up, you know, not checking email. So like one workflow, that's a pretty straightforward workflow. I receive, I think, probably like 200 or 250 emails a day. I find my email to be quite unusable. Most of this is things that actually are not, you know, relevant to my day to day. So if we just go in and say like, you know, hey, what's the latest from email that I need to get back to? Right. We can do this. Simultaneously, we can do concurrent tasks at once.
12:05So if I want to also go ahead and you see from the Alexander Chapman Slack. Can you read all the candidates that they have sent us and make me a dashboard of our candidates that we are speaking to? Great. So this is a strong demo here. You are exporting the entire Slack corpus every night, every week or using the API or you have an agent in that room? How do you get the data out of there into the model? I'm curious. So basically what we do is we run like a cron job at a frequency of about, you know, every 15 minutes. And so every 15 minutes, we'll have Kimmy, for example, go and read everything that I get in my Slack across all of the different channels.
12:57And then it will go ahead and take all of that information. And then we create, you know, we use Chroma, we use Olama, and it'll put that inside of this ChromaDB, right? And now it'll almost have like a recreated knowledge base, if you will, of our Slack. So we'll do things like this. Got it. And is that more stable and a better way to do it than like OpenClaw? With OpenClaw, with Perplexity Computer, with Co-Work, we find it's not comprehensive. It gets confused at times. And in one conversation I had with my OpenClaw, it was like, you should just export the entire database every Saturday night and upload it or something.
13:37And that doesn't seem like a really great strategy. So how do you know it's comprehensive and not so brittle? So this is kind of like the wow moment, like why companies like Deal, why Slash, why Corgi, basically are running almost a ton of different workflows actually on Aragon is because you don't have to, for example, specify. Like I was just, we had to fly to New York. We're doing a really exciting partnership with a very exciting ERP company. And they said, for example, oh, can you make an MSA and can you send it to Josh? And then they realized, they're like, oh, wait, I didn't say Josh from Aragon.
14:18Right. But Aragon actually knew. It sent it to me. Right. And like, they're like, whoa, that's like really, really cool. It's like, actually, it's that good. where like I can even on first time, I can say, hey, just send it to Josh, right? And it by itself, the way we've architected it, it'll go look at calendar, it'll go look at, you know, recent, right? And it will basically say, okay, it's Josh at Aragon is who we're talking about here versus like almost anywhere else. It will ask you and say, oh, which Josh? Here are five different Joshes, for example. Do you, right? Okay, so all of this then gets rolled up into the model on some cadence.
14:54And then, cause this reminds me of like a lot of things I do with OpenClaw or perplexity. But when it's in the model itself, do you have an example where it's in the model and that provides a massive lift for the company? Like what's the most compelling example of that? For example, okay, two things. So one is like, okay, I just went through probably, in this case, like, I don't know, maybe 50 emails or so. So we have to act now. So Vinod Kozla wants to meet, Nikki is his EA, we need to respond back with different, you know, availability, etc. Like, that's a pretty important one, right? So great.
15:34We should probably tell Vinod when we're available to chat about Aragon. But here, for example, Milton is our cleaning service for the office. Can you go ahead and pay the Milton invoice for me, please? And I can have Aragon go and actually pay that invoice. as well. So that speaks to the agentic nature of this and the hooks. You have it hooked into a bunch of different services, I take it. So it's your QuickBooks or your PayPal or your Venmo. Yeah, it also has a financial infrastructure too, Rog. You partner with Slash to allow enterprises to give their agents bank accounts. So this is one of the cases where the agent sort of has a little bit of financial agency.
16:18Yeah, exactly. And so then the thing that we are working on with like our partners is also building out this ecosystem inside of Aragon. So for example, coming very soon, you'll be able to do a backslash, right slash, and then they're, you know, they're, they provide credit cards, they provide, you know, bank accounts, they're like, you know, a Brex competitor, if you will, right. And how do you actually, you know, issue credit cards for these agents? How do you actually set, you know, different policies, different limits, you know, for the cards that you issue these agents, you want to do it in like a secure kind of centralized Sure.
16:52Human in the loop, maybe. Maybe only pay certain known accounts. So this is all part of the blocking and tackling 101 that these agents need to catch up on. So you're really competing directly heads up with OpenClaw, Perplexity Computer, and QuadCowork on the interface for agentic computing. But then on the back end, I keep getting back to the SLM, the small language model you're building? Because that to me seems like there would be some massive lift. So show me something there or unlock that for me, because that's what I haven't seen in the demo yet is that unlock. And by the way, man, this is just another startup I should have invested in.
17:37Oh, wait, I'm on the cap channel of this company. You started, while you're getting cooking there, if I remember correctly, just agents for sales, and you were so far ahead of everybody, right? This is two years ago or something. Agentic Sales was like kind of the first ICP. Yeah. So, I mean, exactly. So, we basically kind of started there. And then pretty much we did like our pre-seed with, you know, A16 and a few others. And what we found when we started showing companies was like everyone wanted to ask for different, I would say, like sources to be able to come into Aragon outside of maybe just sales.
18:15And it became pretty clear to me, like actually, you know, from like a venture model, everyone wanted to verticalize. But like the power of these like frontier models, you know, just kept getting better and better and better and better. and it begs the question of like are we going to have such a thing as like a vertical kind of like a sales you know agent or are these going to be so powerful that you can just create right like a sales agent just from inside of Aragon and have like Aragon be like your source for all sorts of different you know specific agents and then what we decided to do is actually go to a partnership ecosystem approach where for companies that are going really deep so I'll shout out you know Orange Slice is a YC company.
18:56They've done a phenomenal job. They're like a clay alternative. And we have companies like Corgi that basically will do like all of their enrichment natively inside of Aragon. And then that's powered behind the scenes by Orange Slice, right, where they went really deep there. And so then that's kind of the approach that we've been taking as well. So while this cooks, just as far as, you know, the points on this RL piece and updating the weights behind the scenes. So basically, what it's going to do is like, for example, all of your sales, right? If you look at or all of the guests that have been on this week in startups, let's just say.
19:35So these are all facts. This is all historical. This has all already happened. Right. And so like, we just did this pretty fast, I would argue, like, it would be, you know, an order of magnitude faster than anywhere else. And then also on the cost side, right? It'd be much cheaper to do inside of here as well, right? And so here is like for our recruiting, right? Now I have my dashboard here for all of the candidates that we're speaking with, right? Awaiting response, you know, follow-up, source, so we're sourcing, you know, Minx from Stanford, Srinath from AnyScale, et cetera. This is actually all real.
20:11This is all production. We run Aragon inside of Aragon. Regulation and compliance, they're not what anyone gets into starting a company to do. You have a big dream and you've got goals. It can feel like, oh my God, it's getting so cumbersome to do all these chores. This is exactly where a reliable partner like Vanta can help. Vanta automates your entire compliance process through one easy to navigate AI powered platform. And this goes way beyond SOC 2. If you're selling to European customers or enterprises, if you're doing business in regulation heavy states like California, or if you're integrating any kind of protected health information, even if you're just accepting payment card data, Vanta can help you with your paperwork and security so you can focus on making great deals and building a great product.
20:58And that's what it's all about. And Vanta is that great product and they're that great partner for your startup or a large company and big enterprises. Twist listeners get$1 ,000 off at vanta.com slash twist. That's V-A-N-T-A.com slash twist for$1 ,000 off. And because it has in the weights, because it's updated itself, it doesn't need to go for that basically workflow and read all of my email, read all of the Slack. Right. So it should be lightning fast. Exactly. And it should be incorporated into its logic at a more base level. explain to the audience, if they've heard the word weights over and over and over again, but explain to them, given your product, what that means for, let's call it like if Uber had Aragon, they had an SLM, they had the entire history of the company, every quarterly report, every employee had ever been there, every single Slack message for over 10 years, Because every, or they made an S, or maybe they make multiple SLMs.
22:07There's one SLM just for complaints and customer support. One just for, you know, code, et cetera. Because I would assume that at some point you might break it up and have 10 SLMs by vertical so that they were even more tweaked. Am I correct in that? Absolutely correct. Yeah. So you'll actually have multiple. And so what it means for a company like Uber is basically two things, cost and performance. So for the first thing on cost, right? Let's say you have some sort of like workflow that requires knowledge of like all of your customer success, you know, documentation, for example. So instead of basically a harness trying to pull all of this context into a frontier model every single time, right, that info will just be in the weights.
23:01And what that means is much less tokens, like 100 times less tokens will actually have to do that. So therefore, it's way more cost effective. On the second piece, which is like the speed and performance, the workflow will be done much faster, right? Like 10 times faster as well. And so the best way, you know, the audience can kind of think about this is if you go to a frontier model and you ask it, you know, who is Abraham Lincoln, for example? imagine if in order to answer the query a model would have to open up a browser use go search google go pull all of the info yeah we don't have to imagine it because sometimes it does that right it's so sometimes there's some breaking news story you're like tell me about coachella weekend two and madonna's performance and it's like okay searching web and you know that's going to take 60 seconds and that's why you turn notifications on but actually actually so for that, that's a little different, right?
23:59So like, for example, for like the Abraham Lincoln, right, I'm saying it doesn't have to go and do that, right? And the reason why is because it was trained on who Abraham Lincoln is. And so it's in the weight, right? So for a company, right, if you want to find out, for example, you know, like, how did we grow over 2019 to 2024, something like this? All of that's historical, it's already happened, right? Instead of right now, you have to go and you have to read every single record, every earnings, every quarterly statement from those, you know, 2019 to 2024. And then that might take actually two hours, right, for a model to do.
24:40And that might actually cost you like 1500 bucks, right? All of that info should just be in the weights, it should be instant, the model should be trained on this, it should know that, right? Or things like what you were just mentioning, which is like, very recent, you know, for example, like a Coachella, like you should still be able to do tool calls and go search the web for like info that maybe is not inside of the weights of the model. But those things are like, you know, much faster, kind of lower token usage and far, you know, and a little bit more few between, right? So our view is basically, in order to actually own your intelligence to have an asset that the company owns, that's actually something that is theirs.
25:21The only way to actually do this is to train a model on your proprietary data and then give you the weights, right? So you own it. It's yours. And that's what we're working towards. It's genius. What does it cost? How do you charge for it? This is always very important. And it's always a great tell for a founder, for me, Lon, when I'm working with a founder, how crisply they can explain the business model. Here, we'll give it a shot here with Josh, who is a very, I would say very strategic founder in my experience. So we charge by tokens. We have a$5 cost per million tokens blended. And so you can comp that with what Anthropic charges, which is like$15 per million tokens for Opus.
26:03So it's$5 per million tokens. Amazing. And how would that compare to using a frontier model? It's anywhere from, you know, a third to like, you know, half the price. And they pay a service fee as well? Do they have to pay something for this integration? I would think you would need forward deployed engineers type of thing to do this today. So talk to me a little bit about when you have a cutting edge project product like this, the implementation, because if you went into Uber, now it's like, okay, you could boil the ocean here. That company is awash in data. So you probably have to find a department that wants to do it.
26:44You probably have to find a data set. You got to clean a data set. You got to create a process for it being updated. Talk to us a little bit about what Aragon does there. So what we do is we do have a forward deployed motion. And we do start, like to your point, on identifying workflows and use cases where you can start. So I can give an example of Corgi, which is a very fast growing company. They might have announcements of their own. I don't know if I'm coming out about different maybe rounds. So they're a very fast growing company in Silicon Valley. And they do insurance. They're an AI native insurance, basically carrier.
27:25And so what we did is we said, okay, hey, what is your inference costs look like right now that you know from like a frontier perspective shout out to corgi um and basically they told us how much they were spending right on inference basically with frontier and so we said okay look we can actually do this for probably a third of the price like and they're like okay that's very interesting and so then it was like all right let's find a workflow that currently happens it's like very repetitive and manual and let's start there and And so the example of that is we're on their website now. You can actually go and basically request a quote.
28:05And what would happen before is when you go request a quote, a startup puts in different information about their company. So like Aragon, you know, our stage, our funding, you know, revenue, et cetera, et cetera, right? And you'll go and, you know, finish this workflow. Basically, you would put in all of this information. and afterwards what would happen is you know there's a lovely uh person in operations her name is evelyn she would take all this information she would then manually figure out like what coverage you need what policies you need and then they had an algorithm that they would put that in to give you an actual quote right and imagine doing this like 40 times a day that was her whole day And so now what happens is all of that goes into Aragon.
Read the full transcript
28:51Aragon is trained on the policies they have. Aragon is trained on the coverage they have. Aragon is trained on kind of how they underwrite, right? So all of that comes in end to end. And then Evelyn gets a WhatsApp saying, hey, Josh from Aragon just requested a quote. Here's the policies you need. Here's the quote. Do you want me to go publish it to his dashboard? Yes or no? Wow. And all of that is now done. For an accounting firm, keeping your books in order is table stakes. It's the bare minimum. them. It's the baseline. But for an early stage and Series A founder, you need a partner that actually understands the world of startups.
29:24It's very different. And you have to understand the world of venture capital because startups have boards and venture capitals who've invested. And that partner is Pilot. Pilot is the largest accounting firm that's built specifically for startups. You're not just doing your taxes. You're getting CFO-level advice on running your company. And those aren't just words. They have actual former CFOs and other seasoned operators on hand to answer your growth and scaling questions directly. All of their tools are intuitive. They're easy to use. You can track your expenses in real time. And Pilot's going to help you model cash flow scenarios before you make big decisions.
30:00So stay focused on scaling and let Pilot take care of the books. Plus, Twist listeners get$1 ,200 off their first year. Go to pilot.com slash twist to get started. That's pilot.com slash twist. So still human in the loop. Obviously, you're setting a proposal. You want to do that, but it's going to be done faster, better, and cheaper. That's always the trifecta. That wins. And this is amazing. Congratulations. You went through the accelerator a year or two ago. Was it LA34? I'm trying to remember which cohort it was. I think it was like maybe May or something of last year. So it was 34, I think.
30:36Yeah. And yeah, great pivot in there. Some great new investors. you pulled through about half the companies that come through our programs pull through we like to take a lot of risk um and i think i've i always track in our database now where we found somebody you i said if you got rejected from yc please email us we'll we'll take a look at your company and you emailed us and i did uh i'm not dumb uh we met you and we're like this kid's really smart I don't know. I told the team, because I have my notes here, I don't know if he's going to be successful with this startup, but I know Josh is going to be successful.
31:14So let's start the relationship now. Josh, so proud of you. Congratulations. And always let me know when I can be helpful to you. You have my phone number, you got my email. Always let me know if we can be helpful. I'm good for a retweet once in a while or a prescient introduction, but great job, Josh. I really appreciate you. And it's just great to be in business with you. Absolutely, Jason. It's been a pleasure. And for anyone in the audience that's listening, I think Jason and team has built an absolute world-class accelerator program. They've been tremendous, especially right in the early stage.
31:48You really have an idea. And I think a lot of times for anyone in the audience, we just announced our 12 million at 100 seed, but everyone listening, founders, everyone is just trying to figure things out. And I think that the best thing, you know, Jason and team, they really have an amazing, you know, structure and cadence and resources at the earliest stage for how you can like think through your business. And that was helpful for us. And I would encourage everyone listening to absolutely reach out and maybe soon it'll be an agent on Jason's side that. I suppose. Yeah. You cut me off at the pass, brother.
32:27I was going to say, like, we are working on this now because, you know, when you get over 10 ,000 applications, we have a weighted system. Speaking of weights and, you know, Marcus is a really talented young fellow I have working for me and he's been building this system. And what we found was with our AI, our weighted system does essentially the same job a human can do. It can identify that. Now, we still have to decide who do we meet with. And then after we meet with them, we have to make another decision. But what I really want your help with, Josh, this could be a great project for us, is to look back at our meetings and everything, put it into a VSLM, very small language model, on my laptop that I don't train, you know, OpenAI and Anthropic or Gemini on how to do what we do here, our secret sauce.
33:19But then, you know, just be able to say, hey, we missed some things. Because when we miss something, Josh, what I do now is somebody just told me, Molly Wood, who worked for us for a bit and co-hosted with me for a while, she had found a military tech company. And I was like, oh, we don't do that. It's kind of not what we do. But now we do do it. And two years ago, she found this one and it pulled through. And I don't know if it's raising$100 million now. And I was like, okay, if we missed it then, but we built a little relationship, what can we learn? Number one. Number two, hey, let's go to thesyndicate.com and we'll have a phone call with them, invite them on This Week in Startups and see if there is an opportunity for us to do a late stage investment, which the members of our syndicate love to do.
34:01Yeah, there it is. There's thesyndicate.com. All right. We'll talk to you soon, Josh. Let me know if you're ever in Austin and we'll go grab some barbecue with lawn. Any excuse for lawn to get some salt lick with a Terry Black's beef rib. You're not a vegan or something like that, Josh, are you? Please. No, no. We're definitely in for some Terry Blacks. I got the Terry Blacks beef rib and brisket. Very solid. But if you come out to the Hilled Country, I'll take you to the original Salt Lake. You got to see it. It's very Texas. It's pretty Texas. And they got bison ribs there. They got the bison ribs out of the Salt Lake.
34:37I broke a tooth. You know, I had this like 20-year... I'm an old man now. I had a 20-year... Crown? What do they call that? A brew canal. Yeah, a crown. You had a crown. You broke a crown. I had a crown. I'm just chewing on this beef rib. Snap. That's how you know it's a good beef rib. Yeah. When unk loses a tooth, it's a good beef rib. All worth it. I'm getting it put back. It's getting put back. Totally worth it. All right. Good to see you, brother. I wasn't fishing for the endorsement, but I'll clip the heck out of it. We're going to have our agent throw that up on the website right away. Already working on it.
35:08Here's my tweet. Pull up my tweet here. PredictSkills.MD and VSLM. Oh, sure. I'm on it. I'm on it. I've got a typo there are on a collision course and it's going to be insane. Imagine legal, HR, marketing, et cetera, SML. God, I got to get that right. SLM paired with an open claw scale running relentlessly and recursively as he mentioned on your Mac Studio Dell workstation madness. Free, constantly improving intelligence. It's happening, folks. And I put a picture of this Dell NVIDIA GB300 desktop which Michael Dell has not sent me yet. How dare me. But I've been lobbying for. I mean, I'm super excited for this future.
35:48I mean, that's I think to me, that's the crucial step. We've been trying to have the agents run in a sandbox and in Amazon's rack somewhere that they need. They need to spread their wings. They need to get out there in the world. You got to have your Mac Studio or your amazing Dell standalone system. That's where you're in. Second, we can get these, you know, working on a laptop. I don't know the 48 gig is going to do it, but I think it'll be the 128 gig eventually. Going to upgrade everybody to a$5 ,000 laptop from whatever$1 ,800,$2 ,000 one, it'd be$3 ,000. It'll cost me an extra three times 25 people,$75K for the next three years.
36:28So it'll cost me an extra$2K a month if it has a three-year lifespan. But just think about how wild it would be if everybody's laptop was just cranking. I mean, it's also dangerous and scary. It is, but it's that recursive loop. It's that you, every time your agent does something, you're teaching it how to do that thing, and it's refining it over time. That's the crucial thing you miss out on if your agent can't really spread out and make its own sort of way in the world. Yeah. Hey, anything going on in Polymarket? I got a Polymark going for my Knicks tonight, or what's going on? I wanted to, so we should talk about this briefly, that Moonshot has launched Kimmy K26.
37:04This is their most advanced open source model to date. It upgrades for long horizon coding, agent swarms, more proactive agents, claw groups to which you can sort of bring in your own agents and skills from outside. But it did lead me to this poly market, which I will share right now. The Sharps are debating when are we going to get K3? Here we go. So you can see not a lot of volume, only about 30K in volume so far because this is so new. But if you look at June 30th, the Sharps are betting 50. Now it's a 59 % chance that we are going to get Kimmy K3 by the end of June, which would be very exciting.
37:43And we're at 2.6 or 7. 2.6. What was the relief? We were on 2.5. 2.6 got released this morning. So we're just, everybody's just getting their first crack at checking it out. People were kind of speculating here that it's really great at coding. And that's what everybody's focused on right now. Cursor's got a new model coming out. I saw they were raising a$50 billion across X.com. People were talking about that. You got Codex. You got QuadCode. Grok's doing great work there. They'll have some product coming out, I'm sure. It's just over and over and over again, we're going to see everybody get lightning focus.
38:21I heard that Sergey, who I saw Saturday night at the Breakthrough Prize, and I talked to him for a bit. But my understanding is he sent a memo. Is there like a Sergey memo that came out either this weekend or something that got leaked? And, you know, these memos are kind of designed to get leaked, as I've always told everybody. According to the information, the Google co-founder said in a memo to DeepMind employees that, and here's the quote, every Gemini engineer must be forced to use internal agents for complex multi-step tasks. Right. So they're dogfooding. They want them to, you know, catch up here and developers, developers, developers.
39:03Once you get recursive developers going, it informs your whole product production. So people have been talking about, wow, it's incredible how much product is getting shipped out of Claude. Well, Claude focused on code. Code then let them release more product. And then you get this flywheel going. The other companies are obviously trying to catch up with that. Yeah, exactly. Should we bring on our next guest? Absolutely. Let's keep this train of experts only moving here. If you listen to Twist, you're going to get practical and tactical, and you're going to get to peek around the corner with experts only.
39:42So that's what we're doing. He is the founder and CEO of Iona. They make fully integrated autonomous delivery systems for logistics operators. The idea is like a white glove drone service that if you're already doing logistics, you could just add drones to your fleet. Give it up for Etienne Louvet. Etienne, thanks for joining us. Thanks for having me. It's a pleasure. All right. So do you want to demo and show us what you built? Do you want to talk a little bit about why you built it and who the customers are and what it costs? Definitely. I think the team might have the videos in the background.
40:15I got some videos. We can pull up a video right now. Here we go then. And maybe before it even started, I can explain a bit like the origin story because it explains it all. Iona is an island in Scotland, the company. So I'm French. Unfortunately, I can't hide my accent. But we're located in the UK and Ireland as well as in the US now. Actually sponsored today by Gradient, a co-working space in Tulsa. Oh, there it is. We've got a plug in for Gradient. 195 francs per desk. Here we go. Use the promo code TWIST to get six francs or a pain au chocolat. No, wait, wait. It was in France or Scotland?
41:00No, it's in Oklahoma, actually. Yeah, Tulsa. Oklahoma, okay. Shout out to Oklahoma, yes. Beautiful shot video, by the way. It looks like you shot it with a drone. We have multiple. No, what Iona is doing is basically, it's coming from a vision. You can be on the Isle of Iona, which is a remote island in Scotland. You can be an IT consultant with a Starlink, a Zoom, Slack, and so on. But what you can't get access to is physical goods. So what we're building at Iona is the physical internet. It's also close to my heart because a part of my family is coming from rural places. And it's solving this paradox because at the moment, when you have less than 500 people per square mile, for instance, in a location, that's 99 % of the US that's 99 % of the world and yet our systems and logistics are optimized the way we were doing logistics 100 years ago with like a bulky delivery van delivering everything so what we're doing is we're building the full autonomy stack for logistics operators to take that transition which is crucial for them because unfortunately they realize that the big tech are coming and that autonomy is the biggest opportunity for them, but it's also the biggest threat.
42:16So this is, you're talking about the FedEx, UPSs, DHLs of the world. You're going to provide them with the infrastructure. So in other words, like a white labeled zip line in some ways? Yes and no. So yes, in the sense of it's going to be a white label and it's going to be for logistics operators, but we're going to also reduce the threshold. At the moment when a drone is flying in the US without going too much into jargon, it's flying under a regulatory framework that is made for aerospace. It means that it's extremely complex to manage. Part 108 is coming. It's a different regulatory framework.
42:50We have an expertise in that because one of the - Part 108 is the number, did you say? Part 108, yeah. Yeah. And part 108 is coming and it's a different threshold. And we have an expertise in that because we fly autonomously already in Europe and it's going to be the same system in the US. And what it enables is also to open the range of customers to also more local logistics operators, hospital, anybody with a delivery need. Got it. And unpack part 108 for us. We can get, we can double click on, you know, these kind of things here on the show. That's why we're here at Twist. So tell us a little bit about why part 108 is so different and transformative.
43:29Yes, it's a fundamental milestone. So in Europe in 2021, we had a similar framework. For once, Europe was ahead, not for long. And the reality is it just unifies anyone around the way we think about drone delivery and the way we think about drones in the airspace. So what it does is that it's a risk assessment framework. And one of the co-authors of the framework in Europe also inspired some of the rules for the FAA. So what it creates in the world is that we have a common framework, a global consensus on the way drones should be implemented in the airspace, which means that we can finally have Europe, the UK, the US and other part of the world joining around that.
44:20And in more details, what Part 108 does is, first, you don't need to be an aerospace manufacturer with what we call Part 135 to operate. So it reduces the threshold and it creates a framework that is much more adept to a delivery with a drone, for instance. Because if you're not flying a 747, you shouldn't be regulated the same way. And what it does... My understanding is the B-V-L-O-S. I don't know how they pronounce that in the industry. B-V-L-O-S. I don't know. Is that how they pronounce it? Usually we say B-V-L-O-S, yeah. But B-V-L-O-S. B-V-L-O-S in French is a much more gracious way to say it.
45:05But this is beyond, Lon, visual line of sight. Correct. So I think 107, because I've looked at these companies for 10 years, was like, hey, you can fly your drone if you can see it. But once you can't see it, then you get into, I think you tip into 108. Etienne, am I correct? Yeah, that's the distinguishing factor here. It's true. So the BVLOS, it stands for Beyond Vision Line of Sight. So when you can't see a drone, it means that the drone has to be highly automated or autonomous. And the significant change in that is that it means that you can have multiple drones or aircraft supervised by the same team that doesn't have to be on site.
45:48Because if you don't have to see the drone, you can be in Tulsa and operate somewhere else as well. So this is key. He is much greater. So when we contextualize it, this is kind of like going from level two autonomy to level four. Or if you're using autopilot and you've got to keep your eyes on the road, but when you get to FSD, like level four, Waymo, et cetera, you don't need to be paying attention. 107 line, you need a pilot, like a plane, and you need a line of sight. And then 108, you need an operator, which could be a group of people in Manila. You could have 1 ,000 Athena assistants who could be very easily trained in this to know what to do.
46:31They could be anywhere. They could be getting paid$1 an hour. They could be getting paid 30 bucks an hour. It's that level up. I mean, you think of drone operators as having that remote and they're looking and it's like it's got to be above the tree line. And, oh, we lost contact with it and running after it. And it's going from that to like the air traffic control model where you could have a tower that's monitoring all of these drones. And it doesn't have to be like a group of people that are actually watching the drone. It's obviously we're going to have to get there before we're going to have drone deliveries happening sort of en masse.
47:00I mean, I guess my question would be, like, is the goal ultimately just like unsegregated airspace, like drones are out there flying alongside every other kind of aircraft and we're monitoring them in sort of the same ways? Yes, exactly. So what you refer to is what we call the piece is called the detect and avoid. A second piece is called the onman traffic management. I deeply believe that the way we will basically innovate in drones is going to affect general aviation later on. because it's very hard to do unmanned traffic management for aviation. At the moment, 98 % of the control towers in the U.S.
47:37are understaffed. Right. And it's not going to get better, to be honest, because it's a difficult job and because it takes a lot of time to train people. And that's something that we can also do with a drone natively, because it's a bit like an autonomous system doesn't need a GPS. Well, it needs a GPS, but it doesn't need like Google Maps, basically. And it's a bit the same. because the drones are autonomous, they can talk to each other. And because they can talk to each other, the scale in terms of efficiency is drastic. And this is where we go on the vision. And I like this analogy. When you think about it, roads are a Roman technology from 3 ,000 years ago.
48:17And we should build a physical internet that doesn't rely on roads. We should build a physical internet that can transport anything anywhere. And you're building these in Galway, is my understanding. That is a great town. I haven't been there in 25 years, but I drove from Dublin to Galway. It's a college town. Is that right? You're building the drones in Galway? Yeah, we have a Celtic Tiger here. Yeah, we have a manufacturer there. And the reason is Ireland in general is an amazing place for drones. They operate with the same regulatory framework than Europe, which is the one that will be common with the US in terms of the way we think about it.
49:00So great. But they do it extremely fast. And the fast pace of iteration is especially very important. So we can get the paperwork back sometimes in less than 30 days when it takes up to six months in most places. So it's drastically more cost efficient. Now, do you have any partners in logistics yet that you've announced? Yes, we work with a few CMA, CGM in Europe. What is that CMA? CMA is one of the largest shipping companies in the world. They also operate, they have Miami, Singapore and Marseille. And they also acquired a number of companies. And so they have logistics needs. We're working with Welsh Group.
49:44We have an amazing neighbor in Cambridge in the UK, Welsh Group, that is a family-owned business, and it's exactly what we're trying to depict as a vision. They operate a number of deliveries in Cambridgeshire, and we did a demo with them, a public demo. So we demonstrated the tech end-to-end, how to integrate in their system and so on. And we have a few more exciting news coming very soon, including with high-profile partners. Yeah, and for people who don't know, there's a great program, Enterprise Ireland. I'll give them a shout-out. They used to, like, God, they've been around forever, but they used to even buy ads and host startups from Ireland.
50:26Back in my Silicon Alley Reporter days, Enterprise Ireland always was a big supporter of Silicon Alley Reporter. And I think maybe even the early days of this week in startups. And they gave you, I think, Etienne, a couple of million euro investment or grant. How do they do it these days? No, it's a grand project that we have for manufacturing. That's the part in Galway. And the reason is because there is a nice story also. We operate drones as well in Galway area. It's the area, so it's quite difficult to access. It's quite remote, and there are a number of communities there. So we're working on the project to actually connect the islands and connect a number of remote places there and remote communities with Galway and with other cities.
51:14What do people most want to order? Like when you talk to your consumers, you know, you've got people in Galway, they're on farms, they're in rural areas. What do they most want to deliver? Is it just like I need coffee and milk and provisions? Well, I think this is what's interesting is it's not food, right? You guys purposefully are not doing like food delivery. It's like cargo, right? Yes, we're calling that segment the light cargo. So we know very well Mana, for instance, and we know quite well Google Wing and even some people at Zipline. It's another segment. What I usually say is a drone is like a vehicle with wheels between a motorcycle and a 44 tons.
51:51Same philosophy, but not exactly the same tech. So what we do is significantly different. Like we have to get long range. So we go 60 to 120 miles with a significant volume. So we can get up to 20 parcels in the drone and a capacity to independently drop volume. So we can do delivery one, delivery two, and then go back to base, for instance. So what we really the sweet spot for us, and this is also why we're B2B, is whatever needs to be transported. At the moment, in those places, there is a cruel lack of accessibility for day to day items. We're talking about the parcels day to day. And it's something that affects actually even the purchasing power of local communities.
52:36And then we can actually scale that to medical deliveries, even groceries. It's true that the drone is a bit big. So if you want like coffee and milk, might be a bit of another shoot, but we could do it. I got a big totally order. What's the poundage here? How many pounds? Up to 44 pounds. Up to 44. So that's significantly larger than the other folks out there. It's 420, Jason. Our Taco Bell order might get up there. I don't know. We're close. I don't know. That's a fair point. There's a lot of tostadas coming. 40 plus pounds. Yeah, up to four. The first version that we have is - What distance can it do when it's at max load?
53:15At max load, the XL will need to be the one that you use. And in that case, it's going to do 60 to 80 miles. Holy cow. But wait, 60 to 80 miles, which means 20 miles out, 20 miles back with a buffer? No, it's a straight line. So half of that, a bit less than half of that if you want to back and forth. Got it. So 20, 30 miles, no problem. And then you got to have room to come back. Like the FAA makes you keep 50 % in the tank when you land or something like that? Depending on where you operate, but yes, about. And the thing is, it's fully electric. So also what we build is, again, it's the full autonomy stack.
53:51So for instance, we also have demonstrated a prototype for loading and unloading the parcels autonomously. So you can have like an autonomous network. So you can do like delivery point number one. You load a few parcels, you unload a few others. Then you go to a backyard delivery somewhere. Then you go back to another depot. like you can really have a data-driven autonomous logistics network. Yeah. Talk to me about charging. That's my final question there. It would seem to me, you know, we have supercharger stations, right? And you got Zipline in some markets. You've got a bunch of people. Amazon is doing this.
54:25Should there be a common standard, a common infrastructure for charging drones or are we going to just have a proprietary war now where you're all flying drones everywhere and there's no way to stop and recharge. Because if there were charging stations, they would not, I could put one on the ranch here for y 'all. I could give you two acres, cut off a little spot. You come down here, the noise won't be an issue. There's nobody in earshot. And they don't need to be on a highway. They don't need to be on the 405 in LA or 290 here in Austin. You could put these anywhere and you could all be recharging without having to go back to base.
55:04So I think I would slightly disagree with that because I think it's going in the way of innovation. The drones that you're thinking about, especially the zipline and everything, they're mostly multi-copters. So they have a very short range, a different type of technology. My drones, because they're doing logistics, they're extremely energy efficient. And it's not only because it's sustainable or whatever, it's because it's competitive. The more energy efficient you are, the less expensive you are on a unit cost basis. So I need to have a glide ratio. I need to have a wing, for instance. I need to have a sort of things like that.
55:40So when you - What do they call? Because I know there's quadcopters. I know there's fixed wing. What do you call these ones that are the hybrid? Is there a technical industry term? So if it's an hybrid, it's called an hybrid. In that case, like when they have - Okay, so technically it's a hybrid. Got it. Yeah. The technical name, Lon, for a hybrid. Yeah, it's hybrid. Yeah. Yeah. Ords are like what we call the tilt rotor. So it takes off vertically and it transitions to horizontal flight so that we can have the convenience of the vertical takeoff. And the tilt rotor gets you the glide path. And that's what Joby is doing too for humans.
56:15Yeah. Yeah. It's similar to Joby in some way. So we have less propellers and it's, of course, not the same product. So you're bigger than a zipline, smaller than a Joby is the way to think about it. Exactly. Light cargo is like when you're in the middle. Joby is like the new helicopter. Zipline is the new DoorDash. We're like the new delivery van in the sky. I love it. And what weight do you want to get to? Is there a goal that like, you know, you want to get to 150 pounds, 250 pounds, be able to deliver, you know, something of a little more heft? Yeah. You quote the famous philosopher, never say never.
56:48but we're not targeting that segment right away because the majority, like what I like about my segment is that 99 % of the streams are actually in that segment. So I don't want to over-optimize for the one time you need a plasma TV because you can probably wait for a delivery van or any other delivery. Yeah, plasma TV, that's a weird footprint. But yeah, if Lon's getting kettlebells, you know, Lon's getting in shape here. So have that a twist, Noti Gang, if you want to do some buff AI slop of Ron. But if you wanted to deliver a set of kettlebells, you know, it could be 300 pounds worth of kettlebells and just drop it off on Lon's house.
57:29But lower it slowly. Don't just drop it. That's dangerous. Fun story about that. We were supported by a grant and we had to explain why we were buying kettlebells. And I had to promise, no, it's not for personal use. I'm not like just buying a gym with the grant money because we're testing propellers. And if we can't find a way to like leave the table on the ground, it's going to take off. Yeah. Love it. Kettle bells are multi-use. All right. Listen, Ethan, continued success and be safe out there, right? Safety first. Yeah. That's what we always say here at Twist. Or if you're a burning man, there's a famous expression, Lon, safety third.
58:09Oh, no. One of the first Burning Mans I'm at, there's like a bunch of signs. What's first? If safety's third, what's first at Burning Man? I mean, like have a good time, be creative. But they just. Number one, vibes. Literally, these guys have this huge tower where they're doing aerial stuff. Sure, yeah, yeah. And there's a sign. It says safety third. Vibes first. Vibes above all, yeah. We got so much news. We got off-duty. There's a million places we could go. Did you want to talk about the Breakthrough Prize? Oh, yeah. So a friend of mine, Yuri Milner, and his wife Julia do this incredible breakthrough prize every year.
58:44I show some photos here. And I've been invited. And this was the first year I've been able to go. And what's really nice about this is... Look at all these celebrities that were there. Salma Hayek was there. Salma Hayek was there. Naomi Watts was there. Ben Affleck. Look at this. Yeah. I didn't say hi to any of them, but I had a really nice conversation with Gal Gadot. What? Gal Gadot. Yeah, and that was very nice. I spoke to Gal Gadot. I met Rob Lowe. I met Rob Lowe. I traded phone numbers with Rob Lowe. Julia Chastain was at the table next to me. Yes, Olivia Wilde, I saw her there. And then you just had like just every, and Lionel Richie came up and did a version of, he sang We Are the World.
59:32Sure. I sat next to a woman named Frances who's at Caltech who won the Nobel in chemistry. And we had a really nice conversation. I also, at my table, was Darren Aronofsky, the director. Oh, the great filmmaker, yes. Yes, and so I had a nice conversation with Darren because I went down memory lane with Darren. We're off duty here, so I can chew the fat with Lon. I said, hey, Darren, you don't remember this, but we met with Sean, who was in Requiem for a Dream back in the day, and I think he was also in Pi. He was the lead in Pi. I said, and you were doing a film. I had Silicon Island Reporter.
1:00:05I was writing for a paper magazine. I met you guys and we had a drink and then you were lamenting you needed to do screenings for Pi. And it cost like two or three grand to host a screening. So I did one for my magazine and you came and you did drinks after. Oh my God, it's incredible. And we had this like nice little thing and we started talking. He's doing a startup. You know, everybody who's in startups and venture capital wants to do movies. Everybody's in movies wants to do startup. Of course, of course. And so we got to connect. And then we talked about Ikaru, you know, one of my favorite Kurosawa films.
1:00:36I'm watching a documentary on Ikaru, like a random, you know, I go down into the dark web, into like the crevices of, you know, people trading tapes on Reddit and all this craziness. And I look for weird stuff and I found an Ikaru documentary. I don't know who produced it. It's in Japanese. Somebody took the time to put subtitles on it. And they go through that film, which I love. And so what I'll recommend is Ikaru, this great Kurosawa film. It's been overlooked among the great Kurosawis because it's not a samurai movie. People always focus on Yojimbo, Sanjuro, Seven Samurai. They forget about, yeah, Ikiru and High and Low is the other one I really love that is not a samurai.
1:01:16Well, yeah, that was during his High and Low. Yeah, there it is. Yeah, there's Ikiru. This is an incredible film about a guy who gets cancer. This will be my off-duty for today. Sure. And he works in the government for 30 years. He gets cancer. He realizes he's never lived. then he meets a young uh woman and they kind of in the second half of the film he decides i'm going to live a little and he goes out he starts drinking has to have fun he's got uh stomach cancer but he decides he's going to live a little and then i watch this documentary about it and they're like talking about the lighting and they're talking about uh if you if you do the office scene or if you just go into google uh images you can look for the office scene i know the one you're talking about and in icaru the office scenes are incredible because they stacked all these books and all these papers.
1:02:01Jacob got it. So if you look in the background, it's just nonstop. There's a couple of different images here, like right there behind him, you'll see stacks of paper right behind him. The photo, yeah, there it is. And they explained how they wanted to build these like towers of papers and he was approving them. They couldn't find them. So they went into Toho Studios archive of all the receipts and they brought them there with the dust on it because somebody brought them and they took the dust off and cleaned them up. And Kurosawa had a conniption and was like, hey guys, I want them with the dust.
1:02:31I want this to seem like the most oppressive place ever where he's sitting there and people are complaining to him about getting approval. He goes and he decides he wants to do one thing in his life that just counts. One thing to spread joy. I'm going to leave it there. Great film. So I spent a little time talking to Aronofsky about it and we were like the only two people in California. Do you see this film? It means to live. 2022 Living. You should check it out it is a british remake of the same story of icara it starts it starts bill nighy but it's a retelling he has a he's a bureaucrat he learns he has a fatal disease and he decides to live for one day here it came out in 2022 it was at sundance oh wow so anyway i've been having my hollywood moment these days uh two two weekends in a row i was in hollywood yeah wow now with folks uh and so that is my off-duty breakthrough prize just thank you for inviting me but what I noticed, and I was talking to Francis about this incredible Nobel Prize winner, is the pace of science is moving so fast because of AI that all the stuff we talk about here is extraordinary.
1:03:41But the things they're doing in science and the breakthrough they're making, it's essentially we can choose as a society where to point the ai gun right now it's at developers okay fine okay it's that productivity okay great we're going to make new backgrounds in star wars and i was talking to gal gadot about this and she said the movie she's working on instead of spending two or three hundred million on it before promotions pna she said we can now do it for 70 million and everything's done with ai except for the performances except for the writing and she said, I am great with this because, again, we're talking about CGI.
1:04:19Like CGI, obviously you play Wonder Woman, etc. Is this the Bitcoin movie? But you could lower the cost dramatically. What's that? Was this the Bitcoin movie? She's in that Doug Lyman movie that they're talking about a lot about the history of Bitcoin. Oh, I don't know. Yeah. By the way, she's tall. She is Wonder Woman. She is tall. Yeah. And she's got presents, but she's also a lot of fun. Things are moving fast now. The way they're doing it, he's shooting it entirely in rooms on green screen, and then all the backgrounds are going to be filmed in. Okay, so that's the movie. Pull up a story about this.
1:04:52Let me see what you're talking about here. And I asked her, is it Ben Affleck's new technology? Because Ben Affleck was building something like this, and it was no. But she said, I asked her as an actor what the future was, and this is what she was telling me about. And she said, you know, Jason, you know, Jake, the thing here that is super compelling is because, we are, you know, like on this green screen, et cetera, and you strip everything away. She said, I can actually, you can focus on performance. Like it's all performance. And yeah. Yeah, I'm trying to find a good image. They haven't released a lot of good images yet, but it's in a London soundstage.
1:05:33They're shooting the entire thing in a London soundstage. And this has got to be it. It's called - Oh, it is, because she said Casey Affleck. Bitcoin Killing Satoshi, a$70 million feature film shot entirely on a custom gray screen soundstage with AI-generated backgrounds and lighting. So all the backgrounds, all the lighting is all going to be done by AI. The only thing that's real is going to be the actors in front of this gray backdrop. They're shooting this right now. It's very timely. I was just reading about this. Technology. Ask your, do you Claude, if you do Claude's sidebar or your perplexity sidebar, you could ask it what technology platform they're using to accomplish this.
1:06:13It's a company. I already found it. It's a company called Acme AI and FX. They're building a new kind of network or background just to power this project. Okay, let's book them for the show. Let's get them on the pod. It would be really cool to talk to them. Get them on for Friday. I would love to talk about this. So this is definitely the film she was talking about here. her main point is you know we're going to be able i guess this would be the counter to the end of the movie industry the end of the film industry there's going to be many more films made because of this and yeah you know the whole hand wringing okay i get it set designers losing their jobs uh you know fx people maybe you know there's gonna it's gonna be a bummer for a group of people just like it was a bummer when the typing pool went away the messenger pool went because of email and word processing.
1:07:02But at the end of the day, tell more stories. Okay, those people are going to have to adapt and do other things. And there'll be people who do it with sets, of course. But this does seem to me to be a counter to that. And then what if these things could stay in theaters a little bit longer because there was many more of them? Yeah. We could see a kind of resurgence where people might get out of their seats and go to the theater because there's more adult fare. I mean, you know, like everything can't be Snow White, which she was in. Yeah. I mean, if you think about the variety, I think you're absolutely right.
1:07:36I think it's just like we've seen with so many other fields. It's there's always going to be humans in the loop. I can't make movies by itself. It doesn't know how to make a movie that would entertain anybody. You use A.I. to do certain functions and speed things up and make things cheaper to allow the creative people to do more with the creative side. I mean, I wanted to make a horror movie years ago with my friend, and we felt like on the bare minimum, a practical budget just to get the locations, just to make the costumes, just to do the makeup and stuff. It's like 100K. So imagine if we could just step into a room.
1:08:09We'd still have actors. You'd still have makeup artists. You'd still have the bare bones. Yeah, but you might cut it 50K. You might cut it in half. We wouldn't need to take all the actors to some real location. You wouldn't need to pay all of these people like a production designer. And so, yeah, they're on the balance. There's some jobs being taken over by A.I., but it also allows for a whole movie to get made that might otherwise not. And a human movie written by people, starring people, made by people. So we're not talking about it. I think that was always what led people astray was all those tweets where it's like Hollywood is over.
1:08:40Sorry, actors and all that stuff. It's like I don't think A.I. is ever going to replace the creative people. It's going to give them more tools to make more and bigger things. Correct. And so absolutely great for Hollywood. Got anything else off duty here? Oh, my God. So much stuff. I started watching, by the way, the love story. Oh, right. The Ryan Murphy. Yeah, the Ryan Murphy thing. So the wife and I are watching that. We're two episodes in. And, man, it has got me nostalgic for New York in that period. I mentioned on the show before I got to meet JFK Jr. twice. and it is just like in one scene, they're going to the Four Seasons restaurant with the pool in the middle.
1:09:20There's a party there. I remember going there many times. In another one, they were at another place that I've been to a bunch of times. I forgot what the scene was, but I was like immediately like, oh, I've been to that restaurant a bunch of times. That's always the appeal. That's what Ryan Murphy and that crew do so well. Like if you remember that O.J. Simpson show, the American crime story one, it was the same with L.A. in the 90s. They recreated it so carefully carefully that if you were there you were like this is bringing me right back to this like time and place and moment um yes which is part of the fun for sure part of the fun it is definitely and then oh and they were playing bjork and stuff like that you know i've talked about like you know that here before uh but yeah they're just tons of interesting shots of like his loft which i remember was on Walker Street.
1:10:09And then there's a bar, Walker's, Bubby's, the Odeon, Tompkins Square Park, the Guggenheim. What else were they at? So it's just like really Indochine. Like these were all places I went in the early 90s. So I'm just loving it. But the Odeon's still there. Bubby's still there. Walker's still there. So those three are still there. And he's also, you know, at a gym. I'm not sure what gym he's at when he's there. And they also show him. It's all real. They do their research and they get everything sort of just right. And that definitely helps make the show feel more compelling. North Moore Street.
1:10:55That's what it is. Walker's is on the corner of North Moore and I think 6th Avenue. And there's a Calvin Klein's office. And I was literally speaking to David Geffen, the famous music producer, a week or two ago. And we were talking about the film. And he gets a shout out in the first episode that Calvin Klein is, you know, taking a phone call from David Geffen, the music producer. It was really cool. There you go. What else do you got off duty? You must have something to. So anyway, I'm two episodes in. It's great. I think it's great couples watching because it's a love story. Yes. There's a movie on HBO Max that just popped up.
1:11:30It's Brian Fuller, who people remember he did that show Pushing Daisies. He did that show Hannibal that I liked really much, the Hannibal Lecter show. He's got his first feature film out. It's called Dust Bunny. It's on HBO Max right now. Nobody paid attention to this in theaters, but it's really interesting. It's really fun. It's really whimsical. It's kind of a comedy fantasy horror about this. And it's about this young girl who believes in monsters, and she believes there's a monster living under her bed that has eaten her parents. And Mads Mikkelsen plays her neighbor from down the hall who turns out to be some sort of a man of mystery, a mercenary, a hitman of some kind.
1:12:10And so she hires him to kill this monster that she believes is lurking under her bed. Oh, good premise. And it's the question becomes, is something real happening to this girl? Did her parents abandon her? Were they murdered? That's what Mads Mikkelsen is trying to figure out. But she keeps filling in like, no, I'm telling you, it is a actual monster. That becomes the drama of the movie. It's a lot of fun. Really well designed. Very cool. Could I watch this with a 16 year old? Yes. Or two 10 year olds? 10 might be. I mean, if they are mature 10 year olds, it's a little bit of violence. There's a little bit of mayhem.
1:12:45But 1600 percent. The 16 year old, I predict, is going to really enjoy it. All right. Listen, I have been. Thank you to my friends over at Row.co slash twist. Okay, go there, and you get to the front of the line. You get to see if you can get your GLP. Your insurance checker. Yeah, your insurance checker. Yeah, insurance checker. So I have been looking for great sources of protein. I need stuff that's quick and easy, right? You can make a protein shake. You got all these things. Yeah. It takes a lot of work. I do like doing the egg whites once in a while. I love a little brisket. But I have been really trying to focus in on the most bang for the buck in terms of calories to protein.
1:13:23because when you're on a GLP, you can lose a little bit of muscle mass. You're just eating so much less. Not everybody, but some. So I started looking for the best bang for the buck. Egg whites are up there. Greek yogurt's up there. But this one, turns out, tinned fish. Oh. So I have bought like five different tinned fish, and this is the one right now that I'm really feeling. It's called Wild Planet. I think they're like amongst the more, you know, healthy ones or great or sourced really well. And 160 calories for 28 grams of protein. Wow. You're trying to get that 30 grams of protein in the morning.
1:14:04So sometimes I'll just make myself a piece of toast. I get a little bit of carbs. Okay, whatever. It's from my Happy Jane Bakery I love so much here in Austin. But these anchovies, I love anchovies. And you put a little lemon on it. You put some mustard. So toast. You're making toast and then you're putting the anchovies on it? I'll just dump the whole thing with the olive oil on a plate. I eat them up. Oh, wow. I dip a little bit of my olive sourdough in there. And this is very good for you. I suggest you buy a 10-pack, a tin, Lon. And when you're going to make a bad decision, not saying you make bad decisions.
1:14:38I'm not saying you don't. I'm not saying you're overweight, but I'm not saying you couldn't lose 10. I could lose 10. I'm not saying any of those things. I could lose a little bit of weight. I think 10. um so uh this would help you and then i looked it up and now my youtube algorithm is serving me up a bunch of friends of mine like tim ferris was talking about uh people go on anchovy fast so i don't like this fasting it's like i get dizzy i gotta perform on the show i get grumpy but people do anchovy fast now wow so they'll do like three cans of this a day they reduce themselves to 600 800 calories whatever it is uh instead of fasting full-on fasting and this puts you in that ketosis So anyway, listen, I'm no expert on this stuff, but I have lost the 40 pounds.
1:15:22I do feel great. And there is something about a shape I've been since my 20s. There is something about tinned fish, I think, that is sort of unappealing to me. But then I think back, like, as a kid, tuna fish out of the can was one of my favorite snacks. I would just eat the tuna fish right out of the can. So I've developed this over time. I didn't used to have a thing about tinned fish. Okay, I showed you mackerel. The other one that's really good is sardines. Yes. So you can get sardine fillets, and they're incredible as well. Also, same situation. You know, you're going to get into keto. You can also get, and there's people who really love this stuff.
1:15:58Like, you can make a little plate of them. So sardines are probably my, I like mackerel above sardines, but I like, it's just right behind it. So wild sardines, mackerel, and then, I think I tried one other fish. Yeah, no, I think those are the two I've tried. Anyway, people make plates of these. So you make like a charcuterie, but you include some of this. Oh, you know what? White anchovies are the other really good ones. The bocorinos from Spain. And then herring, of course, the classic elderly Jew snack. Well, I want to get, there's a herring from Sweden where they mix it with a mustard cream sauce.
1:16:37Yeah, yeah, yeah, for sure I know about this. And I had it there a couple of times when I was in Stockholm with Tyler and when he was living there. and man that is i'm gonna find that recipe where they do like i think it's like egg whites and mustard and it's a very creamy sauce yeah and they dump the herring in it or they'll put cotton in it's unbelievable so there's your tip folks i hope you stay healthy and we'll see you on wednesday for twist
From the publisher
This Week In Startups is made possible by:
Quo - https://quo.com/TWiST Vanta -
https://vanta.com/twist
Pilot - https://pilot.com/twist Plaud -
https://Plaud.ai/twist
Today’s show:
*Eragon CEO Josh Sirota believes that you should “own your intelligence.” He demos a live enterprise AI operating system connecting email, Slack, calendar, and financial data into a single agentic layer that’s pre-trained to understand any business inside and out.
PLUS we’ve got IONA Drones founder Etienne Louvet, a first look at Moonshot’s Kimi K2.6, and a brand new Off Duty with JCal and Lon.
FULL SHOW NOTES ON NOTION: https://thisweekinstartups-docket.notion.site/TWiST-2278-SHOW-NOTES-34872ca918e380bd9c71e684ce2db7c9?source=copy_link
Guests :
Eragon: https://eragon.ai
Josh Sirota on X: https://x.com/joshua_sirota
IONA: https://ionadrones.com/
Etienne Louvet on X: https://x.com/etienneldva
Relevant Links
Eragon Seed Round coverage: https://techcrunch.com/2026/03/18/this-startup-wants-to-make-enterprise-software-look-more-like-a-prompt/
Neurometric: https://www.neurometric.ai/
Ollama model: https://ollama.com/
Orange Slice: https://www.orangeslice.ai/
Corgi Insurance: https://www.corgi.insure/
Slash Financial: https://www.slash.com/
The Syndicate: https://thesyndicate.com/
FAA Part 108 Explained: https://pilotinstitute.com/part-108-explained/
Zipline: https://www.zipline.com/
Google Wing: https://wing.com/
Joby Aviation: https://www.jobyaviation.com/
Enterprise Ireland: https://www.enterprise-ireland.com/
Éire Composites: https://www.eirecomposites.com/
Kimi K2.6 blog post: https://www.kimi.com/blog/kimi-k2-6
Kimi K2 series on GitHub: https://github.com/moonshotai/Kimi-K2
NVIDIA GB300 NVL72 work station: https://www.nvidia.com/en-us/data-center/gb300-nvl72/
Timestamps:
0:00 Josh from Eragon AI joins the show
2:01 Plaud: If your work depends on conversations — interviews, meetings, calls — you need a Plaud NotePin. You can check it out at https://Plaud.ai/twist and use code TWIST for 10% off!
5:07 Why founders should own and control their data sets
8:33 What is a "company world model?"
9:48 Quo (formerly OpenPhone) - Quo gives you a clean, modern way to handle every customer call, text, and thread all in one place. Try it free at https://quo.com/TWiST
11:32 Eragon demo
15:55 Giving agents their own bank accounts
19:07 Inside Eragon's first partnerships
20:14 Vanta - Get $1000 off your SOC 2 at https://www.vanta.com/twist
25:50 How does Eragon make money?
29:11 Pilot - Visit https://www.pilot.com/twist and get $1,200 off your first year.
39:33 Etienne Louvet of Iona Drones joins the show
42:37 What is FAA's Part 108?
47:13 How innovation in drones impacts the aviation industry
50:26 Working with Enterprise Ireland
58:28 Jason attended the Breakthrough Awards
1:04:10 Doug Liman's AI-powered "Bitcoin" movie
1:09:09 "Love Story" on FX and "Dust Bunny" on HBO Max
1:13:10 Jason's love of tinned fish
Subscribe to the TWiST500 newsletter: https://ticker.thisweekinstartups.com
Check out the TWIST500: https://www.twist500.com
Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp
Follow Lon:
Follow Alex:
LinkedIn: https://www.linkedin.com/in/alexwilhelm
Follow Jason:
LinkedIn: https://www.linkedin.com/in/jasoncalacanis
Check out all our partner offers: https://partners.launch.co/
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




