The machine that explains Lucas Ochoa's career is not the one his company sells. It is a phone charger with a body. At Carnegie Mellon, Ochoa, Gautam Bose and Marisa Lu built Emoto, a multi-axis robotic stand that could turn, lean and gesture. The point was not to make a phone dance for applause. It was to give an invisible intelligence a physical vocabulary. If a machine could move expressively, perhaps a person could read its intention.
Years later, Ochoa is still working on that translation. The stage has changed from a design studio to an office full of portals, PDF files and aging enterprise software. The robot has mostly disappeared from view. Yet the question remains: can a machine understand what a person is trying to accomplish, and can it make its own behavior legible enough to earn trust?
Automat, the San Francisco company Ochoa co-founded with Bose in 2023, is their current answer. Its agents learn from screen recordings, operating procedures and live demonstrations. They read documents, operate interfaces and use APIs. The useful bit is not that they can click a button. Office software has been replaying clicks for decades. Ochoa wants the agent to understand why the button belongs in the process, notice when it moves and recover without summoning a small priesthood of bot mechanics.
“Software should reduce work rather than relocate it.”Lucas Ochoa on Automat’s builder philosophy
A childhood of prototypes, documented online
Ochoa's old portfolio is a cheerful archaeological dig. Its categories are “design + hci + robotics.” The work includes robotic-hand iterations, experimental interfaces, a group Snapchat concept called Portals and Wallet Capo, a compact wallet he put on Kickstarter before starting college. There are traces of a maker who did not wait for a job title to grant permission. Plastic, code, sketches and campaign pages all qualified as ways to think.
At Carnegie Mellon, he found the language and the people for that habit. Ochoa studied design and human-computer interaction, spent an exchange term studying industrial and product design at KAIST, joined hackathons and worked on projects ranging from CMU Hyperloop crowdfunding to Emoto. He has described arriving after high school without a clear sense of his professional tribe. CMU Design, he later said, was unusual because it made him feel accepted for who he was.
That belonging became a partnership. Ochoa and Bose were classmates, collaborators and eventually colleagues at Google. Ochoa has joked about their long list of experiments, including an AI-powered breakfast cereal sorter. The cereal is a funny detail, but it reveals the method: give a messy physical situation to a sensing system, build the interaction around it and see whether the machine can do something intelligible.
The robot moves into the browser
After Carnegie Mellon, Ochoa's résumé moved between interaction design and machines. He worked at Microsoft in user experience and hardware interaction design. At Google, he passed through Creative Lab, led hardware interaction work tied to Pixel Buds Pro and then joined Everyday Robots at Google X. Bose was nearby, applying transformer models across Chrome and cloud experiments. They saw language and vision models getting better at turning instructions into action.
The obvious founder story would have put them into home robotics. Ochoa has said he was obsessed with robots and considered building a company around them. Instead, the pair chose a less photogenic habitat: the computer screen. A software agent needs perception, action and feedback just as a robot does. Its world happens to be made of login forms, loan portals, spreadsheets and shifting web layouts.
This choice also exposed a business problem hiding in plain sight. Enterprise systems frequently do not talk to one another. A document lands in one place. A person reads it, types the useful parts into another system, checks a third screen and hands the work to the next person. People become connective tissue because software integration ends where the actual job begins.
Traditional robotic process automation addressed the gap by recording sequences of clicks and tying them to interface elements. It worked until an application changed. Then the script could not explain what it was trying to do, much less find a new route. Ochoa's critique is both technical and economic: an automation can remove visible labor while quietly creating maintenance labor. The organization still pays. The cost has simply wandered into another department wearing a developer certification.
The designer’s argument for context
Automat calls its answer agentic automation. Customers show the system a process through a recording, a standard operating procedure or a live walkthrough. The platform turns the demonstration into a workflow that combines deterministic steps, where consistency matters, with adaptive steps for unstructured documents or changed interfaces. Forward-deployed engineers remain responsible for keeping the result working.
That last clause matters. The agent industry enjoys a demo in which a cursor buys a cheeseburger. Ochoa prefers the less cinematic test of an automation running 1,400 times a day in a banking environment. Production work demands an audit trail, predictable performance and someone who owns the failure. Automat records sessions and exposes actions and reasoning so an operator can inspect what happened. Adaptability without accountability is improvisation with an invoice attached.
The context layer
A script remembers coordinates. Ochoa argues that a durable agent needs the goal, the meaning of what is on screen and enough judgment to handle change while preserving hard rules where precision is required.
Ochoa's language is revealingly human. Agents learn the way a new hire does. They work in tools people already use. A forward-deployed engineer acts, in his analogy, like a talent-development lead. Trust accumulates through visibility, feedback and shared work. This is not anthropomorphism for decoration. It is a design model for deciding where the person belongs in a system with partial autonomy.
The model also carries Ochoa back to his earliest work. Emoto tried to make a phone's hidden intelligence readable through motion. Automat tries to make an agent's hidden reasoning readable through records, traces and outcomes. One gave the machine a body. The other gives it an audit trail. Both ask the person to understand what the machine thinks it is doing.
“Our automations have a high-level goal in mind and actually understand the ‘why’ of each action they take.”Ochoa, speaking with Carnegie Mellon School of Design
The founder meets his own back office
There is a pleasing irony in Ochoa spending Automat's first two years managing invoices, adding users and issuing company cards. Founders are not exempt from the connective tissue. They merely encounter it at inconvenient hours. He described one banking interface as “navigating an extraterrestrial design,” a line that contains equal measures of design critique and exhausted operator.
As the company grew, an operations assistant and chief of staff took over those routines. Better financial software gave Ochoa visibility without requiring him to check a dashboard ten times a day. “Now I don't need to check the platform daily. I can build the company,” he said. The anecdote is small, specific and more persuasive than a market forecast. Work that recedes gives attention back.
Automat's own growth has raised the stakes. The company announced a $3.75 million seed round in 2023, then a $15.5 million Series A led by Felicis in November 2025, bringing its reported total to $19.25 million. Ochoa and Bose have described serving global enterprises across banking, financial services, insurance and manufacturing. Early access to the Automat platform broadened the pitch from a service delivered with engineers to a place where business users and technical teams can build and monitor agents together.
Ochoa calls the company's development intentionally non-linear. Customer deployments expand while product infrastructure is built underneath them. That sounds untidy because company-building is untidy. His public writing has become more precise as Automat has grown, moving from broad claims about automation toward essays on brittleness, context, trust and the awkward incentives of an old software category.
Cubes, criticism and the quality bar
When Ochoa returned to Carnegie Mellon in 2026 to speak on behalf of Y Combinator, he explained Automat's culture through his education. The literal quality bar, he said, came from the craft and intuition learned by drawing professor Mark Baskinger's cubes. Optimism about technology came from studio courses. Criticality came from lectures that asked what technology does to people as well as for them.
It is an appealing recipe for an AI company because each ingredient corrects the others. Craft without optimism becomes precious. Optimism without criticism becomes salesmanship. Criticism without building produces excellent objections and no object. Ochoa's career has kept all three in circulation, from a wallet small enough to hold in one hand to an agent spread across an enterprise workflow.
His aspiration is correspondingly practical. He wants systems that take on the whole job, including enough understanding to adapt when the environment changes. He wants the effort to leave the customer's side instead of resurfacing as tickets, maintenance or extra headcount. The ambition is large. The proof arrives in deliberately boring units: one less retyped document, one bot that repairs its route, one founder who no longer needs to issue every card.
The future of work is often painted with grand abstractions. Ochoa keeps returning to the screen in front of an operator and the reason behind the next click. It is a designer's scale of attention applied to an enterprise-sized problem. Somewhere between those two dimensions, the robot finally gets to work.