An engineer who retrained as a designer, then went looking for the thing every AI was missing - and found it hiding in your folders.
Open your laptop and stare at the little folder icons. Documents. Downloads. A graveyard called "stuff." That window - Finder on a Mac, File Explorer on Windows - has barely changed since the Reagan administration. Abhay Agarwal looked at it and saw the one place artificial intelligence had quietly skipped.
In November 2025, his company Poly walked out of two years of stealth with $8 million in seed funding and a blunt pitch: the file browser is the most important app you have ignored, and it is time to make it think. Agarwal calls the result "an LLM with infinite context from your life." Investor James Cham of Bloomberg Beta put it more plainly - "we all desperately need the file browser to come back."
Poly is not a chatbot bolted onto a search bar. It is a replacement for the window you use to find things. Drop in terabytes - documents, slide decks, spreadsheets, photos, audio, video, raw code, even saved URLs - and Poly indexes the lot with its own embedding model, Polyembed-v1, built in-house rather than rented from someone else.
Ask a question in plain language and it does not just locate the file. It summarizes, tags, tabulates, transcribes, and can spin your scattered material into a podcast or a presentation. Then it shows its work, citing the exact page number or the precise timecode in a video. "Poly was built on the belief that in an AI-first world, the file system itself has to evolve," Agarwal says. The app shipped first on macOS, with Windows promised next.
The name is not new. Before Poly there was Polytopal, his human-centered AI consultancy. The thread running through both: most software treats your own information as dead weight. Agarwal treats it as the most valuable context window you will ever have.
Most founders pick a lane. Agarwal kept switching them on purpose. He earned a degree in electrical engineering and computer science at Berkeley in 2012 - the hard, mathematical, build-the-machine path. Then he did something engineers rarely do: he went to Stanford's d.school for a master's in product design, voluntarily crossing into the world engineers tend to dismiss.
In between sat Microsoft Research, where he held a fellowship in AI for social impact and worked on computer-vision hardware to help visually impaired people navigate the world. That project lodged a question in his head that he has been answering ever since: AI had no shared design language. Engineers and designers could not even brainstorm as equals. "Show me a company where designers feel equivalent to AI engineers when brainstorming technology," he has said. "It doesn't happen."
So he tried to build the missing vocabulary himself. He named the idea Post-Cognitive Design, taught it to students at Stanford's d.school and later at UT Austin, and started writing a book with the very Agarwal title The Lingua Franca of AI: A Toolkit for Designing Intelligence.
Through Polytopal, his team co-created a dance-choreography algorithm for the VR rhythm game Beat Saber.
They also built a virtual baking assistant for Nestle Toll House that designs a cookie recipe around your dietary needs.
Spotify, Meta and Nestle hired Polytopal to turn messy data into intelligent systems people could actually use.
Poly did not start as a file browser. When it went through Y Combinator in 2022, the product used AI to generate design assets - textures, materials, art for creators who needed them fast. "Poly's focus has always been to empower creators with easier access to design assets," Agarwal said at the time.
Then the ground shifted. As large language models swallowed the creative tooling world, Agarwal pointed the same instinct at a bigger, duller, more universal problem: nobody can find anything they already own. The pivot took the company into two years of quiet head-down work, building an embedding model rather than wrapping someone else's API. When it re-emerged, the framing had grown up. NextView's David Beisel described Poly as "turning everyday files into a personal intelligence layer that helps people work and think faster."
It is a neat closing of a loop. The man who spent years arguing that AI needed a design language built a product whose entire job is to make sense of the unstructured mess humans actually create.
He signs his online life as @Denizen_Kane, a handle that follows him from Twitter/X to Medium and predates every startup he has run. He is an engineer who deliberately went to design school - the academic equivalent of a chef enrolling in farming. And the through-line in his portfolio is gloriously varied: a dance algorithm for a VR game, a cookie-recipe bot, and now a piece of system software meant to outlive the trends that birthed it.
His pet theory, Post-Cognitive Design, is the kind of phrase that could only come from someone who has stood on both sides of the engineer-designer divide and refused to choose. He is, by his own framing, less interested in building one app than in defining the grammar everyone else will use to build theirs.