As developers increasingly lean on AI for code generation, a new perspective from rsbh.dev suggests that the same reliance on AI for technical writing is actively hurting engineering communication. The author argues that while AI-generated code is acceptable because machines parse it efficiently, documentation is fundamentally different because it is consumed by humans. When AI drafts documents, it produces content that is difficult to read, unnecessarily verbose, and plagued by inconsistent terminology across paragraphs.

The Problem with AI-Generated Prose

The core issue identified is that AI lacks an understanding of the reader's flow. Unlike a human engineer who pictures a specific peer with known technical capabilities and vocabulary, AI writes in a 'zig-zag' pattern that ignores the audience. This results in documents filled with jargon and redundant length, where the same concepts are described using different words in different sections. For builders trying to ship fast, this noise creates friction rather than clarity.

Where AI Still Has a Role

The author does not dismiss AI entirely but suggests a stricter division of labor. AI remains valuable for research and gathering raw content for documentation, but the actual writing must be done by humans. This is particularly true for high-stakes documents like RFCs (Request for Comments) and RCAs (Root Cause Analyses). These formats require precise communication to align teams and prevent future outages, tasks that require human empathy and contextual awareness that current LLMs lack.

The Redundancy of Auto-Generated API Docs

A significant portion of the argument targets auto-generated API documentation, which the author labels as largely redundant in the AI era. Traditional auto-docs often amount to little more than tables of JSON field names and types. Since AI agents can directly read source code or parse OpenAPI specs, the value of a human-readable, auto-generated table has diminished. The real need is for humans to write the explanatory layers that connect these technical details to business logic and usage patterns.

Key Takeaways

  • AI-generated code is acceptable for machine consumption, but AI-generated docs fail human readability standards.
  • Human writers are essential for RFCs and RCAs because they understand the specific technical background of their peers.
  • Auto-generated API documentation is becoming redundant as AI can parse OpenAPI specs and source code directly.
  • AI should be used for research and content gathering, but not for finalizing the narrative flow of technical documents.

The Bottom Line

Stop letting AI draft your RFCs. If your documentation isn't readable by a human peer without a headache, you're wasting time shipping noise. Use AI for research, but keep the keyboard in human hands for communication.