When everyone can build more, deciding what should exist becomes increasingly important.
I started as a software engineer in Mumbai, building gesture-based applications with devices like Kinect, Leap Motion, and Google Glass. There, I independently won a project, and it's a fun story how, to build an application for a Scotland-based startup that tracked the hand motions of surgeons on a laparoscopic simulator. I delivered it end to end: PM, developer, tester, and customer success, all in one. It went on to win several accolades, including one at the World Innovation Congress, but its real impact on me was more personal. It's where my love for product building began: anticipating needs people hadn't articulated yet, and the joy of creating something that delivers real value in the world.
That project took me to Duke for my Master's, and from there to Microsoft, where I've spent years building products at the edge of what AI can do, agents included. Shipping systems whose behavior you can't fully predict has changed how I work more than I expected it to. It has also made me more deliberate about the parts of the job no model does for you. A few principles I build by:
Principles
Start with the problem, not the technology.
A new capability tells you what you could build and nothing about what should exist. The problems worth taking are the ones people are already working around badly, the ones that would still matter if the technology went away.
Ship a complete thought, even when it isn't a complete product.
Ship too little and you learn the wrong lesson from how people react. Wait too long and what would have been a delight is already table stakes. AI compresses that window and raises the cost of getting it wrong, because users have been burned by half-finished AI features and extend very little benefit of the doubt. And GA is the midpoint, not the end: most of what you need to know arrives after launch, when people stretch the thing in ways nobody planned for. Time to learn and iterate needs to be baked into roadmaps right from day zero.
You can't test an AI system to certainty, so be exact about intent.
Non-deterministic systems can't be proven complete or correct, which means clarity has to do the work certainty used to: a sharp definition of what the product is for, guardrails, and evaluation that takes failure modes as seriously as success. It also means saying plainly what the system won't do.
Autonomy is a design decision, not a default.
People rely on what they can predict and correct, so every increment of autonomy should have to earn its place. Capability and comprehensibility pull against each other, and a product shouldn't get harder to understand as the model underneath it gets better. Knowing where to keep someone in control is as much of the work as knowing where to take them out of the loop.
And a few I work by
Don't outsource judgment.
AI is very good at finding holes in my thinking and a poor substitute for doing the thinking. Taste and rigor in thinking are what help me differentiate, and I don't outsource those. Sloppiness is visible: in a product, in a document, in an argument. I'm careful about what carries my name.
High agency, low ego.
I fill whatever gap stands between the team and a working product, whether or not it's my job. The second half matters as much as the first: I don't need the idea to be mine, and I'd rather be corrected early than right late. Artificial boundaries are an expensive way to slow down work that users are waiting on.
Optimize for long-term relationships over short-term personal wins.
Some of my greatest joys as a PM have come from people I've now worked with across many years. I don't let a short-term win come at the cost of a long-term relationship. Trust across disciplines is what makes it possible to do harder things together, and it's almost always built well before the moment you need it.
Outside of work, I paint, I write, I read compulsively, and I travel, because keeping those human skills sharp, and ensuring we stay curious, empathetic, and kind, feels more essential now than ever. I've also spent years volunteering as an English teacher, which gave me a lasting conviction: educational access and quality decide too much of what people get to become, and AI is our best shot in generations at closing that gap.
So if you're thinking about what should exist next, especially in AI, agents, human-AI interaction, trust, or education, I'd love to talk. Or we can just discuss art, books, or good tea. Find me on LinkedIn.