First Principles

Episode 2 — Aditya Oberai (Developer Relations Lead, Appwrite)What does it take to lower the bar so far that anyone with an idea can ship, and still trust what they built?Aditya Oberai's whole job is access. The SDKs, the docs, the backend primitives that let someone who's never touched auth or a database ship a real app. In this episode we get into the craft of that: being a technologist first, the line between AI-assisted building and vibe coding, and what he sees in a room full of people with less experience than him that he never gets online.Then we turn to agents. Aditya's recent talk makes an uncomfortable case. Hand an agent the same tools, the same prompt, the same everything, and it still makes different calls run to run. None of the things moving that decision are the tools themselves. I take that as the crack under my own work. My days go into coordination, pushing the power to build outward instead of letting it pool in a few labs. But if we can't even pin down what an agent will do from one model update to the next, what does it mean to hand these things real leverage?A conversation about who gets to build, what breaks when they do, and whether we can trust agents with any of it.

Episode 1 — Jatin Mehrotra (Developer Advocate, AWS)What does it actually take to advocate for technology you've run in production yourself?Jatin Mehrotra spent five years in the trenches of cloud infrastructure, EKS, Terraform, CI/CD, before becoming a Developer Advocate at AWS. In this first episode, we get into what separates advocacy that engineers trust from marketing in a hoodie: writing code samples that don't rot, demoing systems honestly, and knowing what's worth going deep on.Then we turn to agents. Jatin's recent work makes the case that you can safely let an AI agent execute its own code by isolating it in a microVM. We take that as settled and push past it: once containment is solved, what about correctness? An agent can run code that's perfectly sandboxed and still confidently produce the wrong answer. That's the gap I spend my time on at EigenCloud, and it's where this conversation gets interesting.A conversation about building, shipping, and trusting what agents actually do