Watch & Listen: YouTube | Spotify | Apple Podcasts | X (Twitter)
Summary
Software just went through a major abstraction shift.
Tools like Claude Code, Replit and Lovable made it possible for millions of people to build software by describing what they want instead of manually writing every line of code.
Dan Lambert thinks robotics is next.
Peppy is building a prompt-based development layer designed to let someone describe what they want a robot to do, simulate the behavior, generate the underlying code and eventually deploy it to compatible physical hardware.
The opportunity is opening because three things are happening simultaneously.
Robot brains are getting more powerful. Hardware is getting cheaper. AI coding tools are dramatically reducing the amount of specialized knowledge required to build.
Dan already has robots costing less than $5,000 sitting on his desk.
But physical AI creates a constraint software never had.
A hallucinating AI can break a website.
A hallucinating AI controlling a robot can break the robot, destroy something around it or hurt someone.
That means vibe coding for robotics requires an entirely new safety layer between an AI-generated instruction and the physical machine executing it.
And even if that problem is solved, robots still face another enormous bottleneck: data.
LLMs could learn from the enormous amount of text, code and imagery already sitting on the internet.
Robots need data about the physical world: how much force to use, how objects behave when grabbed, what happens when a dog walks into the room, how to recover after dropping something, and millions of other edge cases that are almost impossible to capture entirely in simulation.
That is the sim-to-real gap.
Dan's argument is that the industry cannot simulate its way out of it.
Robots have to enter the real world.
They have to fail.
They have to collect data.
They have to, in Dan's words, get “punched in the face.”
Peppy is betting that once enough inexpensive hardware exists, millions of developers and semi-technical builders will begin experimenting with robots the same way people are experimenting with AI-generated software today.
If that happens, the major opportunity may not only be the company that builds the winning robot.
It could also be the company that builds the software layer people use to tell all those robots what to do.
Why This Matters
Most of the attention in robotics today goes toward the visible layers: humanoid hardware and increasingly capable AI models.
Peppy is attacking the layer between them.
Dan's longer-term vision resembles Android more than a traditional robotics company.
Robot bodies could change every few years. Different manufacturers could win different categories. Hardware could become increasingly commoditized.
But the developer environment, prompts, applications and robot skills could persist across hardware generations.
That creates the possibility of an entirely new software ecosystem built around physical machines.
The crucial difference is that software operating in the physical world has consequences.
Robotics development therefore needs to combine the accessibility of vibe coding with the reliability, simulation and safety constraints required to control machines operating around humans.
That is the problem Peppy is trying to solve.
Key Numbers
Under $5,000
Worst token fraud incident Lenhard says he encountered over a 48-hour period.
10 hours to 10 minutes
Dan says his first experience setting up ROS 2 took more than 10 hours. Peppy wants to reduce that initial onboarding process to roughly 10 minutes.
~$20,000
Dan's example of what a highly dexterous robot hand can still cost today.
~$1 million
The size of Peppy’s recent pre-seed round, according to Dan. He says it was significantly oversubscribed.
~50,000 to millions
Dan’s estimate of the difference between the humanoid robots currently in the world and the scale he believes is necessary for meaningful real-world experimentation.
90%
Peppy’s rapid-prototyping target: get a robotics application roughly 90% of the way there in simulation before finishing the hardest portion with hardware and potentially a robotics engineer.
$2,000-$6,000
The hardware price range Dan wants to see more manufacturers target so far more people can experiment with robots.
December 2026
Peppy’s planned public launch, allowing people to access the prompt interface directly through its website.
Episode Links
Follow
Mentioned
Further Reading
NVIDIA: Jetson Thor and the Computing Platform Powering Physical AI
The hardware Lambert points to as one of the reasons robotics is reaching an inflection point. NVIDIA says Jetson Thor delivers up to 2,070 FP4 TFLOPS of AI compute and 7.5x the performance of Jetson AGX Orin, enabling real-time reasoning and robotics workloads.
Read NVIDIA’s Jetson Thor overview
Figure: Building the Real-World Data Robots Can’t Find on the Internet
Figure’s Index project tackles one of the biggest bottlenecks Lambert describes in the episode: physical-world training data. Figure says more than 16 million videos have already been contributed and it plans to spend more than $1 billion on data and compute over the next 12 months.
Read Figure’s introduction to Index
Peppy’s Pre-Seed Announcement and the “Lovable for Robotics” Thesis
Peppy’s August 2026 funding announcement lays out the company’s plan to turn natural-language prompts into working robotic code across simulation and physical hardware. The round was led by Element Ventures with participation from Rock Creek Capital, Mana Ventures and Colin Gardiner of Yonder.
Read Peppy’s pre-seed announcement
OpenArm: The Open-Source Robot Hardware Peppy Is Building With
Peppy currently supports OpenArm 2.0 as part of its push to make the same robotics software work across simulation and physical machines. OpenArm provides open-source CAD, firmware, control code and simulation tooling for physical AI research and development.
Explore OpenArm
Timestamps
00:00 Why robots, why now?
01:27 Turning 10 hours of robotics setup into 10 minutes
04:18 What is actually holding general-purpose robots back?
06:09 Peppy’s ~$1M pre-seed and the Android-like vision for robotics
09:41 Why robots need real-world failures to get better
11:57 How a plain-English prompt becomes physical robot movement
15:32 What would robotics look like without NVIDIA?
17:59 From experimenting with robots to building Peppy
20:41 “Lovable for Robotics” and the vibe coding moment
23:06 Peppy’s December 2026 public launch
26:38 Should robots be allowed to use knives?
28:30 Should robots do things differently than humans?
31:15 The moment Dan became obsessed with robotics
33:44 What robotics needs next
Go Deeper
Research → (Coming Soon)
Peppy and the Emerging Developer Layer for Physical AI
A deeper look at the robotics software stack, the shortage of real-world training data, the sim-to-real gap, and whether prompt-driven development could become a standard layer across robot hardware.
Read the Full Transcript
The complete edited conversation with Dan Lambert, organized by topic and timestamp.
