LLMs Don’t Code—They Compile Your Prompts

The author argues that AI coding is best understood as a compiler workflow: prompts are code and outputs are compilations, not autonomous programming. English is a flawed medium for specifying software, leading to brittle, non-deterministic results, so perceived gains often don’t translate into real productivity. Instead of hype-driven ‘vibe coding,’ we should treat AI as a tool and focus on building better programming languages, compilers, and libraries.

Key Points

• AI programming should be modeled as a compiler: prompts are code, outputs are compiled results, and iteration is like recompilation.
• English is a poor programming language—imprecise, non-deterministic, and highly non-local—making AI coding workflows brittle.
• Perceived productivity gains often mask real slowdowns; hype around ‘vibe coding’ mirrors past tech bubbles.
• LLMs add value primarily through search, optimization, and pattern reuse; the human remains the programmer.
• AI may eventually replace some programming tasks like compilers and spreadsheets did, but it should be treated as a workflow tool, not an autonomous coder.

Sentiment

The community is divided but leans toward pragmatic centrism. Most commenters agree with the article's core thesis that AI is a tool rather than a replacement for engineering skill, and that the hype cycle is real. However, there is strong pushback against the article's dismissive tone toward AI productivity gains, its reliance on the METR study, and the self-driving car comparison. The dominant sentiment is that both the 'AI will replace programmers' hype and the 'AI is useless' counter-hype are wrong — AI coding tools provide genuine value for specific use cases while falling short of the transformative claims made by their promoters.

In Agreement

• The compiler analogy is a useful mental model — AI assists coding but doesn't autonomously create software, and engineering skill remains essential for guiding it effectively.
• English is a poor programming medium: prompts are imprecise, non-deterministic, and require extensive context specification, making natural language a lossy interface for code generation.
• The hype around AI coding is overblown and risks causing stagnation by discouraging investment in better languages, compilers, and libraries.
• AI-generated code without careful human oversight creates technical debt, with agent-written code tending toward repetitive patterns and low architectural quality.
• AI works best as a tool within well-designed workflows guided by experienced developers, not as a replacement for understanding code and systems.

Opposed

• Many experienced developers report genuine, significant productivity gains — particularly for library discovery, API documentation navigation, boilerplate generation, and working on unfamiliar codebases.
• The METR study is misleading because it tested experts on their own codebases (the population least likely to benefit), and its results don't generalize to typical AI usage scenarios.
• The self-driving car comparison undermines the article's credibility, since autonomous vehicles are actually operating in multiple American cities right now.
• AI enables work that otherwise wouldn't happen — solo developers and hobbyists use it to pursue projects beyond their available focus or time, reframing value beyond pure speed.
• Enterprise and boilerplate-heavy development is where AI excels most, and dismissing this reveals a narrow perspective on what real-world programming actually looks like.
• Geohot's objectivity is questionable given his own commercial interests in tinygrad and history of shifting positions on AI capabilities.