The Untrainable (saranormous.substack.com)

In a thought-provoking analysis, the article “The Untrainable” discusses the evolving landscape of AI-driven software development and the complexities that come with it. As AI models continue to improve, particularly in coding, there's a growing sentiment within the investment community that many companies may struggle to add distinctive value beyond being “thin wrappers” around superior models. Notably, a recent study from MIT revealed that modern coding agents have increased code production by 180%, yet the intricacies of engineering that resist easy measurement highlight that true value often lies in aspects of software development that are less quantifiable and require deeper human insight. Significantly, the piece asserts that while AI capabilities are expanding, the factors of trust, user integration, and domain-specific expertise remain crucial for successful implementation in complex organizational contexts. For instance, models can’t replace the accountability, understanding, and nuanced decision-making inherent in sectors like law or finance, where outcomes are deeply tied to unique user needs and system intricacies. This underscores a shift towards recognizing that the most valuable AI applications will emerge from customized solutions that leverage private data and maintain robust relationships, rather than generic models that commoditize easily measured tasks.