This trend suggests that any work that can be easily measured or verified is destined to be absorbed by frontier models or cheaper, open-source alternatives, driving down the value of generic AI "wrappers." In response, the new competitive advantage for companies is shifting toward "untrainable" value, which consists of private data, human judgment, and organizational trust.
In fields such as law, medicine, and complex enterprise systems, correctness cannot be determined by public leaderboards; instead, it is defined by internal ground truths and long-term reliability.
Success in this landscape requires navigating the "lock and deadbolt" of private environments—securing the necessary permissions, liability sign-offs, and deep-rooted user habits that massive compute alone cannot purchase.
Ultimately, the most defensible AI businesses are those that focus on unglamorous integration and the translation of private reality into actionable outputs.
By establishing the standard for what constitutes "good" work within a specific domain, these companies build moats that are resistant to the scaling of general models.
As general intelligence becomes cheaper and more accessible, the highest value will reside in specialized systems that operate where public benchmarks cannot reach, relying on human-led oversight and exclusive access to proprietary workflows.