Specialized coding models often fail because a developer's workflow isn't just writing code; it's a complex conversation involving brainstorming, compliance, and web research. The best coding assistants are the most generalist models because every complex task has AGI-like qualities.

While an AI agent can find and propose a fix for a specific line of code, it often lacks the context to identify and solve the problem class architecturally across the entire codebase. Expert human engineers remain vital for this higher-level reasoning and pattern recognition.

AI coding tools automate implementation, elevating the developer's role from writing logic to designing systems, reviewing AI-generated code, and making high-level architectural decisions. The focus moves from implementation details to overall system design and strategy.

Karpathy found AI coding agents struggle with genuinely novel projects like his NanoChat repository. Their training on common internet patterns causes them to misunderstand custom implementations and try to force standard, but incorrect, solutions. They are good for autocomplete and boilerplate but not for intellectually intense, frontier work.

As AI agents handle the mechanics of code generation, the primary role of a developer is elevated. The new bottlenecks are not typing speed or syntax, but higher-level cognitive tasks: deciding what to build, designing system architecture, and curating the AI's work.

With AI agents and non-technical staff now generating code, the primary bottleneck has moved from code creation to code review. The human's job is becoming less about writing lines of code and more about applying taste, judgment, and architectural oversight to AI-generated outputs.

AI models excel at coding because correctness is easy to evaluate. Design is harder because "good" is subjective and tied to human taste, making it difficult to create a training feedback loop. Furthermore, design values novelty and cultural context, whereas software engineering prefers established, reliable patterns.

Since coding agents can perform like junior engineers, the value of simply writing code quickly and correctly is diminishing. The new critical skill for engineers is the ability to judge AI-generated code, architect systems, and effectively steer agents to implement a high-level design.

While developers leverage multiple AI agents to achieve massive productivity gains, this velocity can create incomprehensible and tightly coupled software architectures. The antidote is not less AI but more human-led structure, including modularity, rapid feedback loops, and clear specifications.