Prompt: The Minimum Viable Reasoning

Stop chain-of-thought prompting reasoning models. We are making them worse.

Reasoning models (o3, GPT-5 thinking, Claude Sonnet 4.5 thinking, Gemini 3 Pro Thinking, DeepSeek-R1) do their thinking before they answer. When we tell them "think step by step," we overwrite the exact process the model was trained to run internally. Less scaffolding. More problem.

This prompt is the counter-move. Three lines. That is the whole thing. It forces us to state the problem, what success looks like, and which constraints are load-bearing. Then it gets out of the way.

Who it is for Anyone still pasting 2023-era mega-prompts into 2026 reasoning models and wondering why the output feels flatter than the non-reasoning version.

Works with

ChatGPT
Claude
Gemini
DeepSeek
Grok

Prompt / skill content

# The Minimum Viable Reasoning Prompt

PROBLEM:
[State the problem in 1–3 sentences. No persona. No "think step by step." Just the problem.]

SUCCESS LOOKS LIKE:
[What makes an answer correct or useful? Be specific and testable.]

LOAD-BEARING CONSTRAINTS:
- [Constraint that, if violated, makes the answer wrong. Max 3.]
- [Another real constraint, not a preference.]
- [One more if genuinely required.]

Return the answer. Flag any assumption you had to make to proceed.

Where to get it

Read the original resource.