I lost a coin toss in 1984.
It took me four decades to get even.
I grew up in Calcutta (now, Kolkata), building electronic circuits before I had ever touched a computer - in fifth grade, I was already curious about what a few components could be persuaded to do. By middle-school, when global markets opened up again for India, my school acquired a handful of BBC Micros: 64 kilobytes of memory, BASIC, and what felt like an unreasonable amount of possibility.
My friend Ananth Chandrasekharan and I built an expert-system chatbot on one of them. He was a Beatles fan and wanted to call it Billy Shears. I preferred Pink Floyd. We flipped a coin; I lost. Four decades later, I named my own AI agent Arnold Layne, after Pink Floyd's first single. The naming dispute is finally settled, although Ananth may disagree.
The question kept growing.
I kept wondering what happened when we connected more computers. That question took me into distributed systems and networking, including years at Google working on infrastructure that spread computation across clusters. It turns out that persuading a thousand machines to behave like one is a fascinating problem in its own right.
The machines at the edges of networks are changing. AI agents can reason and act, quantum processors depend on entanglement, robots have to sense and move, and thermodynamic and oscillator systems use physical dynamics to compute. Each raises a different research problem, yet I keep returning to the same question: how do we make these capabilities work across many machines? I think of that longer-term ambition as Civilization-Scale Computing. There is plenty we still have to figure out.
At Outshift by Cisco, I work with researchers, engineers, product teams and open-source communities on parts of this problem. AGNTCY is helping agents find and communicate with one another; the Internet of Cognition explores what they need to share to reason together. Our quantum networking work asks how processors and sensors can establish and use entanglement over distance. I like the point in a project when an idea meets a real system, because the system is usually quick to tell you what you missed.
A few decades later.
Arnold is my personal OpenClaw experiment, and in some ways Billy Shears's spiritual successor. We talk on Discord, Telegram, and Slack. We also email each other. Please don't judge.
In 1984, getting a rule-based chatbot to respond convincingly on a BBC Micro felt like magic. Today, models can write software, work through unfamiliar problems and carry on conversations that would have seemed absurdly ambitious back then. We used to write the rules ourselves; now we train systems at a scale that would have been unimaginable on that machine.
My questions have changed along with the technology. I spent years figuring out how computers could share work and operate reliably as a system. Now the computers we connect can reason, remember and make decisions. How do we get them to understand one another well enough to do useful work together?
An agent can write sophisticated code and still misunderstand what another agent is trying to do. Several capable models can agree enthusiastically on the wrong answer. I run into versions of these problems in my own experiments, which is probably why I keep experimenting.
The part that still makes me smile is how familiar the feeling is.
Back in Calcutta, Ananth and I were trying to make a BBC Micro appear to understand us. These days, I am trying to make a collection of much more capable machines understand one another.
Different machines, same fascination.
Away from computers, I travel and photograph places, often following the histories visible in streets, landscapes and architecture. A family travel site, Rowdy Planet, grew out of those trips. I am slowly turning it toward Atlas of Time, a project about the history that survives in the places we walk through.
My working life has taken me through Google, IBM, startups and Cisco. Along the way I earned a Ph.D. in Computer Science from UC Davis and filed more than 80 patents. The work behind those credentials interests me much more than the credentials themselves.
I use this site, and Coherent Cognition, to work through ideas in public: experiments, papers that change my mind, and questions I haven't found satisfactory answers to. I expect some of what I write here to change as I learn more.