Ep 68 - Demo Hell and the Dear Leader Problem

In Ep 68 of Automate It, Stefan and Ilia take on the robotics problem no sensor can fix: corporate politics.But first, the robot invention game: VLAs for tattoo parlors. Stefan wants a model that reads your shoulder like rough terrain and plans the needle path around your freckles. Ilia points out that most VLAs run off a single camera and are happy to be an inch off, which is fine in a gravel pit and less fine on your arm. We also get a crash course in why tattoo ink stays put, and a review of the laser safety standards in Starship Troopers. Working name: the Inkstorm 3000.Then, politics. Org psych has a name for it ("perceptions of organizational politics"), and the research says it's mostly structural. When nobody can define success, whoever controls the definition wins, and robotics barely has metrics. Cue Amazon managers saving their low performers for layoff season, ML breakthroughs that turn out to be OpenCV with a wrapper, and a Soviet tank designer sent to Siberia for complaining about radiator water. The antidote is boring honesty: weekly updates that list every problem, whoever caused it. The best customers love that. Political ones see it as a threat.The advice for engineers: ask your interviewer which production metric matters most. If they can't answer, you're headed for demo hell. And be wary of any company where nobody makes fun of the Dear Leader. At Polymath, how fast a new hire roasts Stefan in public is an onboarding metric.Want to build robots somewhere that ships? We're hiring: https://www.ycombinator.com/companies/polymath-robotics/jobsTheme Song: BIG DATA “Monster (feat. Jamie Lidell)”

Point One is hosting FUSE, a localization conference, on November 19 in San Francisco. Register at fuse.pointonenav.com with code POLYMATHThis week, Ilia and Stefan sit down with Aaron Nathan, CEO of Point One Navigation, to ask why your phone knows which street you’re on but can’t find the door handle of the burrito joint.First, the game: RTK GPS meets prescribed burns. The result is a balloon fleet that spots fires, a drone fleet that sets them, and centimeter accuracy for the napalm. Along the way we cover the coastline paradox, “Mega Idaho,” and why mileage reimbursement is the real business case for precision.Then the real conversation. Aaron explains how RTK works, why the ionosphere is the main villain of satellite navigation, and why a forest fire barely registers (fireworks neither). We get into why Point One owns over 4,000 base stations rather than borrowing anyone’s, and what it’s like pitching investors on 4,000 one-meter property leases. We also cover jamming and spoofing, why the most safety-critical GPS receivers are the oldest and easiest to fool, and what it takes to beam corrections from a geostationary satellite at 4.8 kilobits per second.Ilia then argues that the only defensible data center in space is a small one, and gets fact-checked on the ISS’s power output live on air.Finally, a decade of building a robotics-adjacent company: surviving COVID and the chip shortage, resisting shiny objects, being “the adults in the room,” and why the unbundling of autonomy means the next robot company won’t need 30 billion dollars.

Two episodes in two weeks!This week's robot invention: a cuttlefish-shaped maintenance drone that has to navigate around a featureless alien sphere at a good fraction of light speed, localizing off pulsars alone. Stefan thinks four pulsars gets you within a couple of meters. Ilia brings up the Rayleigh limit, kilohertz flicker rates, and special and general relativity, and lands on the uncomfortable conclusion that the best positioning system for a ship on a known trajectory is a very good clock. Project Stargazer, obviously. Also: why every sci-fi ship where the stars visibly move is breaking the speed of light, and why localization stays unsolved even in the far future.Then the main topic: the numbers behind Agility Robotics' Digit. Agility is SPACing off roughly $1.8M in 2025 revenue, with a prospectus that assumes hundreds of thousands of humanoids in warehouses where the hourly math doesn't beat a human on a bagel and a coffee. From there, a tour through the autonomous trucking SPACs (Aurora at ~1,400x revenue while burning about a billion a year, Kodiak, Plus, Embark, TuSimple) and how they actually turned out. Stefan explains why AI labs get infinite multiples until the moment they start selling something, why VCs writing $25M checks can't possibly be experts in the thing they're funding, and why a pre-product company pitching a TED talk out-raises one with customers.Closing argument: if you're at a well-funded humanoid company and you don't open source key parts of what you build, you're kind of a dick. Most of these codebases will die with the company. Cruise gave us Foxglove. What will you leave behind?Total rosbag t-shirts now available. Profits go to the Open Source Robotics Alliance. Drop a comment for the link.

In Ep 65 of Automate It, Stefan and Ilia ask a deceptively simple question: how hard can it be to put a big red stop button on a robot?But first, they play the Robot Invention Game and draw synthetic training data for offshore wind farms, which somehow turns into humanoid maintenance robots living inside wind turbines, six-armed monkeys that plug themselves into the walls for power, and the unfortunately named “Whirly Gig.”Then, they dig into something slightly more consequential: robot safety. Ilia introduces the Open Source Safety Consortium and Polymath’s new open-source protective stop, designed to safely stop robots anywhere from 20 feet to 3,000 miles away. They get into why traditional e-stops break down for autonomous vehicles and humanoids, why a networked safety button is much harder than it sounds, the black channel principle, and why Polymath is open-sourcing not just the hardware and software, but the certification process itself.Plus: why robots at trade shows with no off switch are terrifying, why safety should work more like encryption and Linux, and why there really is no excuse for an unstoppable robot when the fix can cost less than $200.If you're into robotics, open source, safety systems, offshore wind farms, or just want to know why the emergency stop button is sometimes located directly inside the robot's kill zone, this one's for you.

In Ep 64 of Automate It, Stefan and Ilia are back after seven months in the trenches closing and delivering Polymath's biggest deal yet — and oh boy do they have thoughts.But first, they play the robot invention game, dreaming up a dystopian HOA enforcement drone that patrols your neighborhood for shutter color violations, checks your grass height with lasers, and may or may not have a paintball gun attached. You're welcome, homeowners associations everywhere.Then, the guys catch up on everything that's changed since October: why closing a company changing deal is harder than building the robots, what Claude Code actually is and isn't good for, and why the real industrial revolution (you know, the one Henry Ford started) never made it outdoors. Plus: why fake startups with great fundraising videos keep out-raising real ones, and what it would actually take to run a lights out facility.If you're into robotics, startup war stories, dystopian surveillance tech, or just want to know why Stefan spent 60 hours building a calendaring tool, this one's for you.

In ep 62 of Automate It, Stefan and Ilia sit down with Fred Parietti of Multiply Labs to talk about the wild intersection of robotics and biology – from personalized pills to robots making cell therapies.They dig into why $500K cell therapies still rely on manual, error-prone lab work, how biology’s “black magic” makes automation nearly impossible, and what it takes to build robots that can work in sterile clean rooms. Along the way, Fred explains why the FDA isn’t the villain everyone assumes, why supplements crushed his soul, and how learning from video demos might finally teach robots the unteachable. It’s a journey through biotech’s artisanal chaos, startup pivots, and the strange joy of realizing your robots are better suited for cell therapy than Instagram.If you’re into robotics, biotech, startup war stories, or just want to hear Ilia out himself as an actual cyborg, this one’s for you.

In Ep 62 of Automate It, Stefan and Ilia imagine a fleet of LLM-powered “Robot Ranger 3000” dog bots to track and protect endangered deer, complete with battery swaps, forest-wide power cables, and some very questionable AI decisions.Then, they dig into the real challenges of enterprise robotics — from integrating with decades-old IT systems to why autonomy is the key to efficiency in mining and agriculture. They also share how fundraising doubles as a sales tactic and why trillion dollar industries still fly under Silicon Valley’s radar.

In ep 61 of Automate It, Stefan and Ilia dive headfirst into robot subversion hoodies, self-cleaning backyards, and the philosophical sinkholes of the Three Laws of Robotics. They explore why "don't hurt humans" is harder to code than it sounds, the limits of common sense in machine learning, and how old sci-fi still haunts today’s AI safety debates.Then, things get weird. From fiber optic controlled drones in Ukraine to perimeter defense bots equipped with buzzsaws, the duo brainstorm real (and, our lawyer wants to say, inadvisable) autonomous battlefield tech. But first, they play the robot invention game and accidentally invent a $2,000 robotic hand-on-a-stick startup.If you're into philosophy, robotics, defense tech, and backyard wood chip problems(?), this one’s for you.

On this episode of Automate It, Stefan and Ilia dive into Aurora’s recent announcement about running driver-out trucks on public highways—an undeniably huge technical feat, but one that curiously failed to make a media splash. But first, they play the Robot Invention Game, drawing a wild scenario: applying deep learning to railroads.

This week on Automate It, we're excited to have a special guest!Carnegie Robotics co-founder David LaRose joins the podcast to share his perspective on the evolution of robotics from academic research to rugged real-world systems. He reflects on what’s changed - and what hasn’t - over the last 15 years, from safety systems that still rely on spreadsheets to why so many autonomous vehicle projects have stalled.David also unpacks the real challenges of selling to the DoD, the importance of enabling the next generation of robotics companies with hardened components, and what it’s like crawling through Florida cornfields in the name of agtech R&D.(00:00) Introduction and welcome to the episode(00:32) Robot Invention Game: Foundational Models for Space Exploration(18:30) David’s journey from liberal arts to robotics via IBM and CMU(20:41) Founding Carnegie Robotics(23:55) Why the robotics industry hasn’t scaled as fast as expected(29:18) Safety systems: still slow, manual, and deeply fragmented(35:17) Why big companies (Amazon, Apple) still struggle with robotics(40:05) Demos gone wrong: off-by-one bugs, e-stop failures, and bedbugs(45:38) Working with the DoD: myths, structure, and real challenges(01:08:30) Launching ruggedized autonomy kits and components online(01:16:33) Environmental challenges and thermal testing tales(01:20:42) Advice for roboticists: Embrace collaboration, not just competition(01:23:00) What’s next for Carnegie Robotics: Supporting the enablers