Alejandro
Castellano

Caddi is the company, and Castellano is its co-founder and CEO. The product has a deliberately humble name and an immodest idea behind it: you record yourself doing a repetitive task - client intake, document filing, an invoice, an email you send forty times a week - and Caddi turns that recording into a real, API-driven automation. The team calls it “record-to-code,” or “automation by demonstration.”

The customers are the people the software industry usually forgets: law firms, accountants, financial advisors, consultants. Professional services. The folks whose billable expertise is buried under back-office admin that no enterprise tool was ever quite built for.

What makes it more than a macro recorder is the bet underneath it. Traditional robotic process automation mimics your clicks; the moment a button moves, it breaks. Caddi tries to understand the intent behind what you did, so the automation survives contact with the real, messy world. That distinction is the whole company.

“Automation by demonstration means professionals can train Caddi just like they would a new hire,” Castellano has said. “No coding, no IT team.” The pitch lands because it inverts the usual order of things. The expert is not the engineer in the corner. The expert is you.

There is a reason this is a hard problem and not just a clever demo. The tasks Caddi targets are the ones that resist clean rules - the judgment calls, the exceptions, the “oh, except when the client is in California” footnotes that live in a senior associate's head and nowhere in a manual. Capturing that tacit knowledge by watching someone work, rather than asking them to write it down, is the part that was impossible until recently. It is also the part Castellano spent a master's degree learning how to do.

And the timing is not an accident. Generative AI made it plausible to read a noisy, human demonstration and infer the structured steps underneath it. The operating experience told him where the pain actually lived. Put a model that can understand intent in front of an industry drowning in repeatable admin, and you have a company. The recording is the spec. The professional is the programmer, whether or not they ever realize it.

Before the cap table, there was a kid in Lima, Peru, who wanted to be an inventor. That ambition pointed him, sensibly enough, at electrical engineering. It is a tidy origin for someone who now sells a machine that learns by watching.

He did not wait for permission to start building. While still in college he launched an EdTech venture called A Plus Tutoring - his first company, at around nineteen. It was small. It was also the thing that introduced him to business people, which turned out to matter more than the revenue.

That introduction led to a family office, where Castellano spent years managing a portfolio of companies and investments. He calls it his “on-the-streets MBA.” It is the part of the resume that does not photograph well but explains everything: he watched, up close and across many industries, how a business grinds its way from one million in revenue to ten - and exactly how much of that grind is repetitive admin that nobody enjoys and everybody pays for.

The childhood dream had unfinished business. He came to the United States for a master's at Cornell Tech, where he went deep on computer vision and machine learning with an entrepreneurial bent. Two halves of a person were now in the same room: the operator who had seen the busywork, and the engineer who could finally teach a computer to see it too.

He landed at the AI2 Incubator in Seattle - the startup studio spun out of the Allen Institute for AI - as an entrepreneur-in-residence. There, the operator and the engineer shook hands, and Caddi was the result.