March 18, 2026
Spec Wars: Return of the Code
A sufficiently detailed spec is code
Engineers say specs are code in disguise; AI fans say bots fill the blanks
TLDR: A viral post argues that today’s AI “code from a spec” demos rely on specs so detailed they’re basically code, pointing at OpenAI’s Symphony. Commenters split: critics say specs aren’t cheaper or clearer, while optimists claim AI can fill gaps—cue jokes about inventing “LLMSpeak” to boss bots around.
The internet erupted after a post claimed the “spec = code” crowd has a point, using OpenAI’s Symphony and its super-detailed SPEC.md as Exhibit A. The doc reads like step-by-step instructions, not a high-level wish list—and that’s the drama. Skeptics say if you have to spell out every field, retry rule, and timer, you’re basically writing the program. So where’s the magic?
Key Points
- •The article critiques claims that AI-driven 'agentic coding' can generate software from specifications alone.
- •It identifies two misconceptions: that specifications are simpler to produce than code, and that specs inherently improve software quality.
- •A concrete example is OpenAI’s Symphony project, said to be generated from a SPEC.md file.
- •The SPEC.md is described as pseudocode with implementation-level details such as database schema, concurrency control, and retry/backoff logic.
- •The author argues that sufficiently detailed specifications effectively become code, undermining claims of simple spec-to-code generation.