The VC's Lens: How AI Is Rewriting the Rules of Defensibility

Every company building AI is asking: if the models keep getting better and anyone can access them, what actually makes us defensible? In this episode of Talking AI, Matt Paige sits down with Avi Bharadwaj, an Investment Director at Intel Capital who writes checks into the software infrastructure layer of AI. Avi breaks down five specific layers where defensibility is actually showing up as frontier models commoditize intelligence, explains why building for the gap between what models can and can't do is a losing strategy, and shares how he uses Claude Cowork agents to scan Hacker News and PitchBook overnight. They also dig into why the chatbot era was brief and agents are now first-class citizens, what world models and emergent abilities might unlock beyond language, and the honest truth about what it's like to be a VC in the middle of the AI revolution. Watch or listen to the full episode: https://hatchworks.com/talking-ai/ --- Key Moments: 0:00 — Intro: The moat question 1:41 — "Better models don't kill moats" 3:25 — Workflow and integration moats 6:27 — Jack Dorsey's "Hierarchy to Intelligence" thesis 9:55 — From data scientist to AI investor 13:12 — How a VC uses Claude Cowork agents daily 18:58 — Specialized models vs. the ever-shrinking gap 22:45 — "Build things that improve as the model improves" 25:43 — Agents are first-class citizens now 31:41 — "Don't use models like if-else loops" 35:08 — World models and emergent abilities 38:34 — When will robots hit mainstream? 43:10 — Coolest and most overrated things about VC --- Key Links: Intel Capital: https://www.intelcapital.com/ Connect with Avi on LinkedIn: https://www.linkedin.com/in/aravindbharadwaj/