Josh Bleecher Snyder, founder of exe.dev, has concluded that Anthropic's Claude operates fundamentally differently from traditional compilers — and performs better because of it.

In a detailed blog post, Snyder argues that while compilers translate source code to machine code within a single layer, Claude works vertically across the entire software stack. The AI can handle strategy, product planning, architecture, and implementation without the organizational overhead that typically separates these functions.

Why vertical integration matters

Snyder illustrates this with exe.dev's distributed DNS server project. The company needed to reduce latency for users outside Oregon and increase uptime resilience for their fast-starting VMs with custom domain names.

Rather than coordinating across multiple specialists, Snyder used Claude to research DNS system designs, explore security considerations, evaluate implementation strategies, and build complete systems with tests. Multiple agent loops worked concurrently, raising questions spanning from high-level architecture to line-level code details.

The AI agents made numerous undiscussed decisions that varied significantly between implementations. For database rollbacks, different agents created wildly different solutions — one using timeline fields to detect history alterations, others taking entirely different approaches.

"It was shocking how many important decisions the agents never asked about but simply made — and made differently," Snyder wrote.

Traditional software development separates these concerns across roles: executives handle vision, product managers create specifications, architects design systems, engineers write code, and compilers generate binaries. Each handoff requires meetings, documentation, and coordination overhead.

Claude bypasses this by operating across all layers simultaneously. While it may not match dedicated human experts in individual tasks, it eliminates the communication friction that typically slows cross-functional work.

Snyder's conclusion challenges the common framing of coding AI as simply "natural language to code" translation. Instead, he positions it as a fundamentally different approach to software development that collapses traditional organizational boundaries.

The exe.dev team ultimately built their distributed DNS server using insights from multiple AI-generated implementations, comparing approaches and codifying successful decisions into written guidance for future development.