Welcome, humans.
Most industrial robots still learn jobs the same basic way: move to point A, save it, move to point B, save it, repeat. That works great until the box moves, the lighting changes, or somebody puts the part in the wrong corner.
Standard Bots CEO Evan Beard is building around the opposite idea: show a robot what you want, let it learn from demonstration, and give normal factory teams tools they can actually operate. He says the company is already deployed across hundreds of companies and roughly 350 use cases.
Industrial robotics, meet the “why does this thing still need a specialist?” era.
Here’s our favorite parts:
(07:48) Vibe coding, but for robots: Standard Bots can turn a natural-language request into a robot program, which is already a big usability jump over legacy interfaces.
(08:35) The bigger trick is demonstration: A person can teleoperate the robot, use gloves, or guide a handheld device, and the model learns the task from examples instead of a hand-coded motion script.
(23:20) The “shoe on the conveyor belt” problem: Real factories produce weird edge cases. Their system keeps collecting corrections so future models can handle more of the long tail.
(30:04) Production-ready does not mean perfect: If a human only has to intervene once an hour instead of standing at a station all day, that can still be a huge ROI. If a mistake destroys a $2,000 part, the bar changes.
(44:34) Evan’s robotics prediction: He thinks U.S. manufacturing could go from deploying tens of thousands of robots a year to millions, then tens of millions.
The thread through all of this is simple: AI matters most where old robots were too brittle to justify the hassle.
Why watch this? Because Evan gives a very practical answer to “what does AI actually change in robotics?” It is not a humanoid doing backflips. It is a factory robot that can handle a shifted part, learn from examples, and be operated by the people already doing the work.
Watch and/or Listen now: YouTube | Spotify | Apple Podcasts
P.S. At 38:35, Evan invent the “Home Alone eval” for humanoids: send one upstairs while an eight-year-old tries to trip it. I would watch this benchmark.
Keep scrolling for a plain-English breakdown of what “learning by demonstration” actually means, plus the last two podcast episodes worth your time.
Additional Resources: What “show the robot” actually means
Traditional industrial robots are basically excellent replay machines. AI changes the setup by letting the robot learn a broader pattern instead of memorizing one exact trajectory.
You demonstrate the task. That could mean driving the robot with a gamepad, physically guiding it, or using another teleoperation setup.
The model sees more than coordinates. Camera views, the robot’s position, and the state of its gripper all become inputs.
It predicts a short trajectory. Instead of choosing one next point, the system can predict the next chunk of motion and keep updating as new visual information arrives.
Variation becomes the point. If different can shapes, part positions, or lighting conditions appear in training, the model can learn to handle those differences instead of requiring a perfect grid.
Human corrections become new data. When the robot hits a weird case, the fix can feed the next training cycle.
That does not mean AI should control everything. Evan explicitly says if a motion is identical all day with no meaningful variability, replaying the programmed motion is still fine. The win is expanding automation into all the messy jobs that never fit the old model. See Standard Bots here.
UPCOMING EVENT: The Neuron IRL in San Francisco
The Neuron is going IRL in San Francisco on November 18 starting at 4:30 PM PT. If you’re in town, come hang out with us for drinks, bites, and a live recording of The Neuron podcast. We’ll have The Neuron community together in one room, plus some special guests and plenty of AI conversation.
Huge thanks to Slack, Alumni Ventures, and Sonar for helping us bring the night to life. It’s free, but space is limited, so RSVP and save your spot.
🎙️ In Case You Missed It…
Two recent interviews and episodes we think you’ll love.
1. Can Living Neurons Make AI 5x Faster?
TL;DW: Alex Ksendzovsky and Jon Pomeraniec explain how The Biological Computing Company grows roughly 150,000 living neurons on an array of 4,096 electrodes to study how brain cells process information. They translate what they learn into software that runs on ordinary GPUs, reporting about 5x faster video generation with roughly 80% lower compute costs. No wet brain required in your laptop.
Why you should watch: Find out how a lab full of living brain cells could make the AI hardware you already use more efficient. The distinction between experimenting with neurons and shipping neuron-inspired software makes this much more practical than the sci-fi headline.
Watch / Listen: YouTube | Spotify | Apple Podcasts
2. What If Apps Didn’t Need Code?
TL;DW: Runway co-CEO Anastasis Germanidis explains Solaris, an experimental interface world model that generates software screens frame by frame instead of rendering a traditional webpage from HTML and CSS. It reacts to clicks, drags, and typing, hinting at a future where interfaces can change on the fly and similar models could power games and robots.
Why you should watch: What does a “world model” actually do? Anastasis breaks down how AI can predict what a digital environment should look like after your next action, and why that could change the way we build and use software.
Watch / Listen: YouTube | Spotify | Apple Podcasts
Have a topic you want to learn about? Request it here!
What do you want to learn about AI?Pick your favorite, then share any others in the "additional feedback" |
|
Subscribe to our YouTube Channel for more!
Subscribe on YouTube to help us bring in more builders, researchers, and guests who can teach you something useful about AI every week.
Stay curious,
The Neuron Team
What'd you think of this podcast episode?Pick an answer below, then tell us why with the "additional feedback" option. |
