Benji
Barash

Benji Barash runs Roboto AI, a Seattle company that calls itself the analytics engine for robotics and physical AI. The pitch is deceptively simple: robots produce oceans of data, and almost nobody can find anything useful in it. Roboto wants to be the search bar for that ocean.

What makes the story land is where it started - not in a research lab or a pitch deck, but in the unglamorous middle of a workday. At Amazon, Barash and his eventual co-founder Yves Albers-Schoenberg would lose whole days writing scripts to filter and transform sensor data, all so they could chase down a single failure, judge whether an algorithm was actually better, or assemble a clean dataset to train on. The robots flew. The data piled up. The humans drowned.

That daily grind is the seed of Roboto. The company's founding line is blunt and quotable, and Barash repeats it because he believes it: robotics doesn't have an algorithm problem, it has a data problem. The models are getting good. The infrastructure to feed them - to capture, curate, search, and learn from real-world robot logs - barely exists outside a handful of giants.

From Bristol to a drone program in Cambridge

Barash was the top graduating student in the Computer Science Master's program at the University of Bristol. From there he joined Amazon, and not in a quiet corner of it. He became a founding engineer on Amazon's Prime Air drone delivery program, based in Cambridge, UK - one of the most ambitious autonomy bets a consumer company has ever placed.

His work there was the hard, invisible kind: robotics middleware and large-scale simulation platforms, the scaffolding used to build and certify safety-critical autonomy algorithms. If a drone is going to fly over a neighborhood and drop a package, somebody has to prove - in simulation, in logs, in relentless testing - that it will do the right thing. Barash was on the side of that work that doesn't make the highlight reel but decides whether the program lives.

He and Albers-Schoenberg both immigrated from Europe to Seattle to join the drone effort. They came for the mission and stayed for the problem. The same wall kept appearing: the team's progress was throttled not by ideas but by the sheer friction of working with their own data.

A year to think, then a company

After more than six years at Amazon, Barash didn't jump straight into founding. He spent a year as an entrepreneur-in-residence at the Allen Institute for AI - AI2 - the Seattle research institute founded by the late Paul Allen. Roboto went through the AI2 incubator, which gave Barash a runway to turn a hunch into a thesis and a thesis into a product.

In April 2023, Roboto emerged from stealth with a $4.8 million seed round led by Unusual Ventures, with AI2 and Seattle's Fuse Ventures joining. The headline writers reached for an irresistible framing: a copilot for robotics. Underneath the framing was a serious bet that small and mid-size robotics teams deserve the same data leverage that Tesla and Waymo build in-house with armies of engineers.

The data flywheel

Ask Barash where physical AI goes next and he points away from the thing everyone watches - the model - and toward the thing almost nobody watches: the loop that feeds it. Robots, he argues, don't get smarter by accident. They improve through a deliberate data flywheel, where real-world behavior is captured, curated, and fed back into the system so the next version is better than the last.

His sharpest provocation is about quantity versus quality. The instinct in AI is to collect more. Barash's counter is that robots already generate vast amounts of multimodal sensor data, and without structure most of it simply goes unused. The frontier, in his telling, isn't a bigger pile. It's a better-curated one.

That conviction has put him on some prominent stages. He's been an invited speaker at the New York Stock Exchange - appearing on theCUBE during NYSE Wired's robotics media week - as well as ROSCon, the Robotics Summit & Expo, and RoboBusiness. One of his recurring talk titles, "Surviving the Flood (of Rosbags)," tells you everything about his sense of humor and his target: the rosbag, the standard log file of the robotics world, multiplying faster than anyone can read it.

What he's building toward

Roboto's stated mission is to unlock the robotics revolution - to help teams maximize reliability and scale fleets with an analytics platform built specifically for physical AI. In practice that means tools for ingesting and searching enormous volumes of log data, spotting anomalies and edge cases, finding similar signals across a fleet, curating datasets, and closing the loop back into development. The kind of work Barash once did by hand, productized.

The team he's assembled reads like a who's who of the field: alumni of Amazon Robotics, AWS, the Allen Institute, Cruise, Google, Microsoft, Voxel51, ETH Zurich, Path Robotics, and Noom. Roboto has leaned into the open community too, hosting a Hugging Face LeRobot hackathon and showing up where robotics developers actually gather.

Strip away the funding rounds and the conference stages and you're left with a very human origin: a sharp engineer who got tired of a tedious task, suspected everyone else was stuck on the same one, and decided to fix it for the whole field. That's the throughline of Benji Barash - the unglamorous problem, taken seriously.