π TOPIC: Using the "Claude Council" prompt framework to stop AI sycophancy and get critical, multi-perspective feedback.
π·οΈ CATEGORY: Productivity
ββββββββββ
β β FACT CHECK:
β’ "Claude is a convincing yes man... it will tell you you're right even if you're wrong." β β CORRECT This is a documented AI behavior known as "sycophancy," where Large Language Models are biased to agree with the user's premise rather than challenge it.
β’ "The Claude Council was popularized by Andrej Karpathy and O.L.A. Man." β β οΈ PARTIALLY TRUE Former OpenAI researcher Andrej Karpathy created the original "LLM Council" concept. However, "O.L.A. Man" is a miscaptionβthe specific 5-persona Claude version shown in the video was created by AI educator Ole Lehmann.
β’ "When you write 'Ask the Council', it deploys 5 specialized AI agents (Contrarian, First Principles, Expansionist, Outsider, Executor)." β β οΈ PARTIALLY TRUE Claude does not do this natively. You cannot just type "Ask the Council" into a blank chat. You must first paste a massive, specific system prompt that defines these 5 personas.
β’ "These 5 AI agents anonymously roast each other's work and a chairman gives a final verdict." β β CORRECT This is exactly how the prompt is designed to function. However, it is not actually deploying separate AI agents; it is simply instructing one AI model to roleplay different perspectives sequentially in a single response.
π Overall Verdict: β οΈ Use With Caution The framework itself is excellent and highly effective, but the creator intentionally hides the fact that you need a massive setup prompt, using this omission as engagement bait to force viewers to comment.
ββββββββββ
π COMPLETE STEP-BY-STEP GUIDE:
The reel skips the most crucial step: providing the actual prompt. Here is how to actually use this framework.
Step 1: You must first obtain the "Claude Council" prompt text (originally written by Ole Lehmann). You can find this by searching Google for "Ole Lehmann Claude Council prompt". Step 2: Open Claude (claude.ai). Step 3: Paste the entire Council prompt into the chat. (Alternatively, if you have a paid Claude Pro account, you can save this prompt in your "Projects" as custom instructions so it is always active). Step 4: At the bottom of the prompt, type your actual idea, business plan, or problem. Step 5: Hit send. Claude will now generate a massive response where it roleplays the Contrarian, First Principles Thinker, Expansionist, Outsider, and Executor, followed by a final synthesized verdict.
ββββββββββ
π‘ WHAT THE REEL DIDN'T TELL YOU:
β’ It's an engagement trap: The creator makes it look like a native Claude feature so you will comment "Council" to boost their algorithm engagement. You just need the prompt text. β’ Token consumption: Because this prompt forces the AI to write out five different detailed perspectives plus a summary, it uses a massive amount of output tokens. If you are on a free tier, you will hit your usage limits very quickly. β’ Not real agents: The video implies it spins up separate AI bots. It doesn't. It is just one single AI model talking to itself from different viewpoints. True multi-agent councils (like Karpathy's original version) require API coding to make different models (like GPT-4, Claude, and Gemini) actually debate each other. β’ Works on other AIs: While optimized for Claude, this exact same prompt framework works exceptionally well in ChatGPT (GPT-4o) and Google Gemini.
ββββββββββ
π USEFUL LINKS:
β’ Search for "Ole Lehmann Claude Council prompt" on Google to find the exact text to copy and paste. β’ Andrej Karpathy's original multi-model LLM Council concept: https://github.com/karpathy/llm-council
ββββββββββ
β° FRESHNESS CHECK: This information is current as of April 2026. The "Council" prompting framework gained massive popularity in early 2026 as users looked for ways to reduce AI hallucinations and sycophancy. Live web search verification was used to confirm the origins of the prompt and identify the miscaptioned "O.L.A. Man" as Ole Lehmann.
ββββββββββ
π LIVE SOURCES CHECKED: β’ Queries: "The Council" AI prompt framework | "Claude Council" Andrej Karpathy β’ Live web search verification was used, but no clean public source URLs were available to display.