Domain Expertise vs. AI: The Moat Debate Heats Up

A post titled 'Domain expertise has always been the real moat' is generating real pushback on HN. The argument is that knowing a specific field deeply is what protects builders from AI commoditization, since coding is increasingly table stakes. Commenters are split: some agree that shipping a spreadsheet is as good as shipping code if it solves the problem, while skeptics point out that models are already pre-trained on most domain implementations, meaning the 'novel' vertical knowledge you think you have may already exist in the weights.

The pattern here: this debate keeps resurfacing because it is genuinely unresolved. Builders want to believe their hard-won industry knowledge is defensible. But the honest counterargument, that historical domain implementations are already in training data, is uncomfortable and not easily dismissed.

The nuance is that the most defensible position is probably proprietary data and relationships within a domain, not just knowing how the domain works. Knowing how insurance underwriting works is not a moat if GPT-4 also knows how insurance underwriting works.