5 Secret Claude Codes (run before every project)
You open Claude, type “build me a landing page,” and fifteen minutes later you have something polished that solves the wrong problem. The fix is not faster prompting. It is five short phrases you paste before you let the model run. They slow Claude down on purpose so you spend less time building the wrong thing.
Most people treat AI like a code generator: jump straight to output. I run a tiny team with AI in the loop, so “wrong direction, fast” is expensive. These phrases are not magic words. They are borrowed from strategy, product, and decision-making work. Each one changes what Claude optimizes for before a single line of spec or copy gets locked in. I am Mika, building with two people and AI on purpose, so when I say these save hours, I mean hours I would have spent unbuilding the wrong thing.
FIRST PRINCIPLES
Aristotle’s idea; people like Musk and Thiel popularized it in business. Use it when you are entering a space with an existing playbook. You tell Claude to strip borrowed assumptions and reason up from what must be true, not from what everyone else already built.
Before we design anything, reason from first principles: what are we assuming that might be wrong?JOBS TO BE DONE
Clayton Christensen’s framework. Use it before you get attached to your solution. It shifts the question from “what are we building?” to “what would someone hire this to do?” That reframe alone has killed bad products before they ever shipped.
Use jobs to be done: list what people are hiring this for, rank the jobs, and stress-test whether my idea is the best way to get those jobs done.SWOT
Came out of business schools in the 1960s. Use it before you commit. It puts internal strengths and weaknesses on the same page as external opportunities and threats, so you are not just optimistic. You are honest about whether now is even the right time.
Run a SWOT on this decision, then say clearly what it implies for go vs wait.STEELMAN
From rationalist and debate communities. Use it before you pitch or invest. Instead of attacking your own idea, you build the strongest version of the opposing view. You actually understand what you are up against before you are in too deep.
Steelman the strongest argument against this plan. Then tell me what would change my mind.10-10-10
Suzy Welch’s framework. Use it for any go / no-go decision. You ask how you will feel in ten minutes, ten months, and ten years. It separates decisions driven by anxiety from ones driven by real conviction.
Walk me through 10-10-10 for this decision: how will I feel in 10 minutes, 10 months, and 10 years?Key insight: The best pre-project prompts do not ask for output. They ask for the right frame so the output stays worth building.
How to use these codes
In Claude or any other LLM, use the same pattern every time: name the code, then paste the thought, topic, or decision you want it applied to. You are not unlocking a hidden Claude feature. You are giving the model a familiar lens (strategy, product, debate, or decision-making) so it reaches for the right kind of reasoning first.
These words are not proprietary to one app. They show up in books, case studies, and everyday business language, so models already know what “SWOT” or “jobs to be done” is supposed to do. Your job is to pair the label with your real context so the output lands on your problem, not a generic lecture.
Here are some related guides to check out: