The Emergence of Coding as a Killer App

Benedict Evans argues that coding has become the first undeniable use case for AI because it is a natural, deterministic fit. Just as the earliest PC users primarily used their machines to build more computing tools, software developers are using LLMs to build software. This is not a coincidence; it is a logical progression of platform shifts. While other AI applications remain experimental or niche, agentic coding has achieved clear product-market fit because the demand from developers is immediate and high-utility.

The Infrastructure vs. Application Value Trap

Evans draws a critical parallel between the current AI boom and the history of mobile data. He notes that while mobile network operators built massive, sophisticated global infrastructure that changed the world, they failed to capture the value themselves. Instead, the value migrated "up the stack" to applications and operating systems.

He poses a central, unresolved question for the AI industry: Will foundation models become commodity infrastructure sold at marginal cost, or will they function like operating systems that capture significant value? He remains skeptical that a chatbot or a raw foundation model is a "product." Instead, he suggests that the true value will be found in domain-specific applications that integrate AI to solve concrete business problems, such as cash flow forecasting for commodities companies, rather than generic productivity tasks.

Evans emphasizes that we are currently in the "early, messy" phase of a platform shift. He compares the current state of AI to the internet in the late 1990s or the PC era of the early 1980s—a time characterized by excitement, technical instability, and a lack of clear winners. He pushes back against the demand for deterministic predictions, arguing that the nature of this cycle is defined by multiple potential paths.

He notes that the current "pricing crunch"—where users are shocked by high token costs or conversely, massive value delivered for low subscription fees—is a direct echo of the early days of mobile data, where pricing models had to scramble to align with underlying costs and perceived value.

The Future of Software Engineering

While the industry is currently fixated on how AI will change the role of junior vs. senior engineers, Evans suggests it is too early to know the outcome. He points out that we are currently automating tasks that were previously done by humans, which will inevitably change team structures and career paths, but he warns against over-interpreting current trends as permanent shifts. The market structure for software development will likely take several years to settle as the supply-demand imbalance in AI capacity resolves.