The Commodity and the Moneymaker (theaiunderwriter.substack.com)

This piece argues a practical buy-vs-build framework for the GenAI era: rent the commodities and build the moneymaker. Because modern LLMs and "agentic" systems are novel, brittle and stochastic, the author recommends outsourcing repeatable, non-differentiated "verbs" (extract, compare, summarize) to vendors that act like picks-and-shovels providers. Important vendor criteria: short-term, non-invasive contracts; reliance on established LLMs/cloud infra; transparency and explainability; verifiable outputs (recognizing true epistemic certainty doesn’t yet exist); and technical vetting by an informed engineer. Red flags include vendors who ask customers to co-build, opaque “secret sauce,” heavy proprietary data reformatting, or shaky 3rd‑party dependencies. Conversely, companies should internally build their unique "moneymaker"—the domain expertise, models and workflows that create sustainable alpha. Shared or vendor-supplied alpha decays fast and isn’t a defensible moat. The post emphasizes practical engineering risks (model verification strategies, feedforward/backprop tradeoffs, precision vs accuracy, overfitting/attention sinks) and urges optionality: short deals, easy extrication, and internal ability to audit model efficacy. Bottom line: rent commoditized components to move fast, but own the differentiated IP that sustains competitive advantage while actively managing vendor lock-in and model uncertainty.