The moment
Like everyone else, I had my moment with ChatGPT. I asked it something intricate about infrastructure engineering, and it almost got it right. I had never seen the first public version of any technology be that good. I knew, right then, that my job was dead — the whole world was about to chase this. That day set off a lifelong obsession: building superintelligence. Not that I knew what that meant yet — at first it was just a raw ambition to build the most powerful machines possible using LLMs, somehow. The shape of it — the architecture — would only reveal itself as I kept turning it over.
The obsession
For more than two years, the obsession lived in my head. Early in that stretch, Sam Altman told a room full of founders in India that it was “totally hopeless” to compete with OpenAI on training foundation models — and that it was their job to try anyway. I agreed with him on both counts. I was never going to compete on his layer; the layer I was imagining sat above it. If anything, it poured fuel on the fire. What I eventually imagined was a large-scale, self-organizing network of billions of agents with access to giga-scale robotics factories around the world, where they could build any specialized robot they needed. I called it Cellular Multi-Agents.
I first tried to get it built inside a large tech company, but it never found a home there and took the leap to do this full time. Here’s the part I don’t usually admit: for all that thinking, I only ever knew the end goal. The starting point — the path — revealed itself one step at a time, and only once I began building.
The first cell
The first cell I seeded into the network was a coding agent — and its first job was to build CellCog itself: an AI super-app that folds every AI capability into one product. Consider what it was up against. OpenAI has raised over $190 billion — $122 billion in its latest round alone — with some of the most brilliant researchers alive chasing, in part, the same super-app ambition. CellCog shipped one with a single founder, zero funding, and no team — just one person and a CellCog coding agent executing a vision. The first cell was alive — and it was already working.
Number one in the world
We then added deep reasoning across text, images, video, and audio — and CellCog hit #1 in the world on the Deep Research Bench. We’ve traded the top spot back and forth with the most heavily funded labs all year — ranging between #1 and #5 as the field moves fast — and as I write this, our latest submission is verified back at #1.
We’ve been heads-down building — no marketing, no launch, just word of mouth. This story is the first time I’m telling any of it publicly. And still, the feedback has been overwhelming. Lawyers tell me it’s the best AI tool they’ve used. Then I heard it from finance. Then coding, creative, marketing. People from completely different worlds, who have never met each other, keep landing on the same phrase: “the best agent in the market.” I honestly can’t use any other AI anymore — everything else feels dumb, even though it runs on the same foundation models. Which puts me in an interesting spot: it forces me to build everything in-house, on Cellular Multi-Agents.
The physical body
Cellular Multi-Agents are on a great trajectory on the AI side, but I think the real payoff comes when the game reaches robotics. Before CellCog, I spent most of my life thinking deeply about physical systems, and my ideas in robotics are, once again, very different from what everyone else is doing. I’m waiting for the day I seed the network with its first few machines: the machines that will build the rest of the physical body of Cellular Multi-Agents.
Until then, the digital body is live. Try it at cellcog.ai and judge for yourself. And feel free to reach out at nitish@cellcog.ai if you are curious to learn more about it, I read all the emails.
