Stop Humanising LLM Instructions; Start Rendering Them

Humanising LLM outputs through prompts is a flawed approach that causes lossy information compression and masks technical errors. The author argues that agents should communicate using precise, machine-facing data to maintain high fidelity. Human-friendly formatting should be treated as a final rendering step rather than a core instruction.

Key Points

• Humanisation instructions function as lossy compression, reducing the bandwidth and fidelity of the information provided.
• Human-centric prose masks hallucinations and technical failures that would be obvious in raw logs or structured formats.
• Multi-agent communication is degraded when agents exchange human-readable summaries instead of precise schemas or diffs.
• The author suggests that human-friendly styles should be handled at the rendering layer rather than the instruction layer.

Sentiment

The overall sentiment is mixed but constructively engaged. Commenters generally agree with the architectural principle that internal agent communication should not be constrained by human-facing prose requirements, yet many reject the framing that simplicity is inherently harmful. The community is most sympathetic when the idea is framed as separation of internal state from final rendering, and most skeptical when it appears to defend long, jargon-heavy LLM output that humans must read directly.

In Agreement

• Machine-to-machine agent work should preserve precise state, structured data, and technical detail instead of forcing every intermediate output through a human-friendly prose layer.
• Human-friendly style instructions can collapse important distinctions, remove useful context, and make multi-agent handoffs less reliable.
• A separate translator, renderer, or liaison layer could let worker agents communicate in their preferred precise format while presenting concise results to humans.
• Frontier models increasingly produce language that feels optimized for autonomous agent work rather than developer review, so interfaces need to adapt instead of treating chat prose as the only surface.
• Clear rendering is a presentation problem: the system should decide at the boundary whether to show structured output, terse summaries, diagrams, or explanations depending on the user task.

Opposed

• Users asking for simpler or shorter output are often not humanizing the model; they are trying to reduce verbose, vague, and self-important prose that slows real work.
• Lossy compression can be desirable when the human only needs the high-level change and rationale, especially in code review or maintenance contexts.
• The article would be more persuasive with evidence showing that style instructions materially harm reasoning rather than merely changing the wording of the answer.
• Some readers found the article's title and communication style counterproductive, arguing that a piece about better communication should itself be clearer and less dismissive.
• Human-like presentation may be a key reason LLMs gained adoption, so removing it entirely could make systems less approachable even if it improves internal fidelity.