For vertical AI applications, foundation models are now sufficiently intelligent. The primary challenge is no longer model capability but building the surrounding software infrastructure—tools, UIs, and workflows—that lets models perform useful work reliably and trustworthily.

While many new AI tools excel at generating prototypes, a significant gap remains to make them production-ready. The key business opportunity and competitive moat lie in closing this gap—turning a generated concept into a full-stack, on-brand, deployable application. This is the 'last mile' problem.

The current ease of delegating tasks to AI with a single sentence is a temporary phenomenon. As users tackle more complex systems, the real work will involve maintaining detailed specifications and high-level architectural guides to ensure the AI agent stays on track, making prompting a more rigorous discipline.

Andrej Karpathy's experience building a 'MenuGen' app, only to see its function replicated by a single prompt to a newer AI model, suggests the trend of AI-assisted app development is a temporary phase. As models get more capable, the need to build a separate application wrapper diminishes.

The accessibility of 'vibe coding' tools enables non-technical builders to create apps. However, they often pitch ideas that the underlying frontier models (like Claude or ChatGPT) can already perform natively within a single chat thread. This creates a wave of redundant software that doesn't need to exist as a standalone application.

As AI makes the act of writing code a commodity, the primary challenge is no longer execution but discovery. The most valuable work becomes prototyping and exploring to determine *what* should be built, increasing the strategic importance of the design function.

The focus in AI has shifted from crafting the perfect prompt (prompt engineering) to providing the right information (context engineering), and now to building the entire operational environment—tooling, systems, and access—that enables a model to perform complex tasks. This new paradigm is called harness engineering.

Developing LLM applications requires solving for three infinite variables: how information is represented, which tools the model can access, and the prompt itself. This makes the process less like engineering and more like an art, where intuition guides you to a local maxima rather than a single optimal solution.

The current trend of using AI to code simple apps ('vibe coding') is a temporary bridge technology. As foundation models become more capable ('Software 3.0'), the need to build and deploy separate applications will diminish. Users will accomplish the same tasks with a single prompt, making many vibe-coded apps obsolete.