Sebastian Peralta

There is a version of the robotics story everyone already knows: a humanoid demo, a viral video, a promise that the machines are close. Sebastian Peralta's story starts somewhere else. A kid obsessed with Einstein and Feynman, convinced he would either invent something that mattered or solve the hardest open problem in physics. He did neither. He built something more useful.

The Physics Kid

Peralta is first generation, the son of an Ecuadorian mother and a Nicaraguan father, born in Boston into a childhood that never stayed still: Wisconsin, then Singapore for middle school, then Chicago for high school. Somewhere in that shuffle he fell for physics, hard enough to build a major and a life plan around it: invent something that mattered, or find the equation that ties quantum mechanics to gravity - the holy grail problem.

Then he ran into large language models and computer science, and the plan splintered. Rather than pick one path, he triple majored in electrical engineering, computer science, and physics, trying to hold all three at once. A deep learning and robotics master's at the University of Pennsylvania's GRASP Lab followed, and that same year he incorporated a company with a mission that still sounds almost naive in its ambition: bring embodied artificial general intelligence to factories, to warehouses, eventually to people's homes.

Student loans got in the way first. He took a job at Google, knowing it was temporary. When ChatGPT launched, he quit, raised a pre-seed round, and started Mbodi AI.

The Problem Hiding in Plain Sight

Give Peralta 25 seconds and this is the pitch: current robots cannot handle high variability, and that single limitation locks roughly $500 billion across manufacturing, logistics, and pharma labs out of automation. You have seen the lab demos. You have not seen a robot actually run a warehouse floor, because the ones already there are manually programmed, can do exactly one thing, and don't do that one thing especially well. The customer bends to the robot. Not the other way around.

Mbodi's answer: software that plugs into any robot hardware and lets an ordinary factory worker, not a robotics engineer, teach it a new skill using natural language and demonstration. The skill is compiled and executed deterministically at runtime. No reprogramming cycle. No six-month integration project. Show it once, tell it what you want, and it runs.

The Pivot Nobody Warns You About

Peralta's first instinct was to build open source libraries for other robotics companies, elegant abstractions he assumed the industry would adopt the way software teams adopt a framework. It didn't happen. Roboticists, he learned, don't pay for software that way. They build in house.

The people who did want what he'd built, deep expertise in generative AI, deep learning, and physics applied to robots, turned out to be the end customers themselves: CPGs, third-party logistics providers, manufacturers, pharma labs. Mbodi stopped selling a developer tool and started selling a complete product to the operators running the floor. The pivot, in his words, changed everything.

Who Buys This, and How

Mbodi's customers are manufacturing, pharma, and logistics companies wrestling with high-variable kitting: picking and placing objects that never come in the same shape or size twice, vials, consumer packaged goods, mixed cases. It is work that has resisted automation precisely because no two shifts look the same.

The go-to-market runs two tracks. On partnerships, Mbodi signed a joint commercialization agreement with ABB after winning the Global AI Startup Challenge out of 200 companies, a deal that has already produced one Fortune 100 customer with more in the pipeline. Mbodi also ran a proof of concept with a Fortune 100 CPG company that has since moved into active pilots, alongside conversations with Fortune 100 pharma labs evaluating pilots of their own. On direct sales, the method is unglamorous by design: get a tour of a mid-size to large warehouse through a warm connection, qualify it, start a pilot.

Betting Against the Black Box

The center of gravity in embodied AI right now is the end-to-end foundation model: point a system at a task and let it figure out full dexterous manipulation on its own. Peralta is betting against that path. His view: embodied general intelligence gets built incrementally, starting with objects that vary only slightly, proving it in production, then stacking more complex skills on top as the system accumulates real deployment data. Stay incremental, and you're live, generating revenue, and collecting the kind of production data a lab demo never produces, all at once.

Where Mbodi Goes From Here

In a year, Peralta expects Mbodi running on roughly 170 robots and at $10 million in annualized revenue, building on the ABB partnership and the pilots already underway. The longer arc is bigger: Mbodi sitting at the intersection of the application layer and live production data, hardware agnostic enough to run on any robot and absorb whatever foundation models come next, functioning as the physical AI operator for an entire industry. Pull it off, and integration time for a new robotic use case drops from six to nine months down to under ten minutes.

Why This Matters

For founders: Peralta's pivot is a useful data point for anyone building deep tech with a developer-tool instinct. The market that wants your abstraction and the market that will pay for your product are not always the same market. The second one is often hiding in an industry you never planned to sell into.

For investors: Mbodi's incremental thesis is a direct counterpoint to the end-to-end foundation model bet that dominates embodied AI funding right now. A company in production, billing Fortune 100 customers, and compounding real-world data today is a different risk profile than one waiting on a generalist model to arrive. Worth underwriting on its own terms.

Sebastian Peralta is the co-founder of Mbodi, an AI software company that lets any robot learn new skills through natural language and demonstration. Mbodi has a joint commercialization agreement with ABB and is currently in pilots with Fortune 100 CPG and pharma companies.

FoundersBrief covers the founders building the physical world's AI layer: robotics, deep tech, and hard tech. This is where we believe the next decade of venture-scale outcomes gets built, and it's where we're pointed. To partner with us or feature your founding story, visit foundersbrief.vc/partner.