Nous Research co-founder Karan Malhotra built Hermes, currently ranked the number one personal agent and AI app on OpenRouter. This interview covers the technical reasoning behind why Hermes improves with continued use, a claim worth scrutinizing closely because most deployed models do not adapt post-deployment.
The conversation goes deeper than a product pitch. Malhotra makes a direct argument for why open-source AI must win, which puts him in direct opposition to the closed-model strategies of OpenAI and Anthropic. The live demo showing Hermes modifying a childhood game is the kind of concrete capability display that reveals more about agent architecture than any spec sheet.
Read this if you want to understand how a small research outfit competes at the top of a leaderboard dominated by better-funded labs, and what self-improving agent design actually looks like in practice, not in theory.
[WATCH ON YOUTUBE →]