Naval, Guillermo Rauch, Blake Scholl, and Max Hodak on Software Factories and Token Efficiency
Naval Ravikant hosts Guillermo Rauch (Vercel), Blake Scholl (Boom Supersonic), and Max Hodak (Science) — three founders building physical or infrastructure products with AI. The conversation centres on how they measure AI ROI, the shift from shipping output to building factories that produce output, and Naval’s operating principle: waste tokens, save time.
Key ideas
- Software factories. Guillermo Rauch’s frame: the engineer’s job has shifted from shipping output B to building the factory that produces outputs B through Z. The question is no longer “how good is this engineer at shipping?” but “how good are they at building the factory?” This implies 100x–1000x leverage gaps between engineers, not 10x.
- Token leaderboards are the new lines-of-code metric. Just as measuring lines of code is a poor proxy for engineering value, measuring token consumption conflates cost with output. What matters is shipped product, not tokens burned.
- Model quality mirrors user quality. Max Hodak’s observation: AI tools reflect the judgment of the person prompting them. Capable engineers get more from models because they give better feedback. This will diminish as models improve, but is significant today.
- Waste tokens, save time. Naval’s operating principle: models are cheaper than human time by orders of magnitude. Don’t optimise token usage; optimise your time. Throw Codex, Claude, and Gemini at the same problem in parallel, take the best result.
- No marginal cost of trying. The economic case for Naval’s approach: the cost of trying multiple AI paths in parallel is negligible; the cost of spending time on the wrong one is not.