{"slug": "56", "title": "56", "summary": "A developer's prompt engineering experiment found that a 56,000-token prompt filled with banned phrases, structure bans, and meta-rules produced sterile, committee-like output, leading them to abandon rule-heavy prompts in favor of a few high-quality worked examples. The author, writing on Tanya N.'s AI tool field notes site, argues that cutting prompt bloat improves output quality more than upgrading to a more expensive model tier.", "body_md": "# 56\n\n## The anatomy of a bloated prompt\n\nIf you are doing a deep dive into prompt engineering, you'll recognize this pattern. My prompt eventually became a graveyard of:\n\n**Banned phrases:** A long list of specific words the AI wasn't allowed to use.**Structure bans:** Rules against specific sentence shapes that felt \"too AI.\"**Opener restrictions:** A massive list of forbidden ways to start a paragraph.**Register rules:** Complex instructions on formality levels for different content types.**Meta-rules:** Instructions telling the AI which rule to follow when two other rules conflicted.\n\nThe moment you need \"rules for your rules,\" you've stopped writing a prompt and started writing a buggy program that no one is actually debugging.\n\n## Over-engineering with validators\n\nBecause prompt rules are requests rather than hard guarantees, I started adding post-generation validators in the code. I built a stack of regex checks to catch banned shapes, vocabulary blocklists, and rejection passes. At one point, I had seven different validators running in a bundle.\n\nI thought I was being rigorous, but I was actually just compounding patches to compensate for the fact that the model was being steered into a corner.\n\n## The \"Committee\" Effect\n\nThe result wasn't a perfect voice; it was a sterile one. The output became flat and cautious. It sounded like it had been written by a corporate committee trying to avoid any possible mistake rather than a human with a point of view.\n\nFor a long time, I blamed the LLM. I actually started preparing a budget to move to a more expensive, \"smarter\" model tier, thinking that would solve the quality dip. I was wrong. A more powerful model given 56,000 tokens of contradictory restrictions would have just been a more expensive version of the same bland output.\n\nIf you're building an AI workflow, stop adding \"don't do X\" rules. Instead, focus on a few high-quality worked examples. Here is a simplified version of how I shifted from a rule-heavy prompt to an example-driven one:\n\n```\n# Role\nYou are a sharp, opinionated tech analyst. \n\n# Voice Guidelines\n- Be concise. \n- Avoid corporate jargon.\n- Lead with the conclusion.\n\n# Examples of Target Voice\nExample 1: [Insert high-quality sample here]\nExample 2: [Insert high-quality sample here]\n\n# Task\nRewrite the following input to match the style of the examples above.\n```\n\nCutting the bloat usually does more for the output quality than upgrading the model ever will.\n\n[Next Prose is the actual control plane in LLM agents →](/en/threads/6303/)\n\n[these AI tool field notes](https://tanyan888.com/), with plenty of directly applicable cases.", "url": "https://wpnews.pro/news/56", "canonical_source": "https://promptcube3.com/en/threads/6328/", "published_at": "2026-08-14 21:53:15+00:00", "updated_at": "2026-08-14 22:28:50.204655+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-tools", "ai-products"], "entities": ["Tanya N."], "alternates": {"html": "https://wpnews.pro/news/56", "markdown": "https://wpnews.pro/news/56.md", "text": "https://wpnews.pro/news/56.txt", "jsonld": "https://wpnews.pro/news/56.jsonld"}}