The Golden Age of AI Applications

The article argues that the golden age of AI applications is defined by three converging signals: the Fable retraction exposing regulatory risk, Satya Nadella‘s thesis that the moat cannot be the model itself, and Salesforce‘s $3.6b acquisition of Fin validating the market. Together, they reveal that building AI applications is not a simple extension of SaaS—it demands new disciplines that go beyond engineering scale or uptime. The core tension is that success now depends on how much intelligence can be squeezed from a finite token budget, which changes what it means to build a defensible product.

The concrete path forward lies in mastering three technical disciplines: picking the right model for each task (e.g., Kimi K2.6 is fast but imprecise, Qwen 3.6 27b is legendary but finicky, GLM 5.1 excels at coding but is slow), designing a hill-climbing loop that lets an agentic system improve over time, and continuously evaluating the model-plus-loop performance. These are novel challenges because the infrastructure and models are moving fast, and each company must define its own loop design—a problem-definition exercise akin to systems engineering, referencing Donella Meadows‘ work.

The serious takeaway is that no single company can staff a team for every workflow; the nuances of tuning these complex engines are better left to a few vendors who can amortize costs across a broad population. Mastery of model selection, loop design, and evaluation—not model size or hype—will determine who owns this golden age. The article makes a pragmatic case for treating AI applications as a systems integration problem, not a race for the biggest model.