You know the feeling. You've got 5 minutes before a meeting. Palms sweaty. You open chatGPT to ask that burning question. Just give some context, a prompt. You wait. It thinks. Another prompt. Not quite. It's missing this and that. You end up more confused than you started. Your self-doubt increases as more avenues open up before you that you thought you were past.
I've done this too. Many hours later and you've gotten 4 right answers - each one perfect at the time until a crack in the shell was pried open with just one more pull of the slot machine. You've chased several mirages as a desperate desert wanderer, only to realize you're way off the path you started on. You loop back to the first solution - "you're absolutely right, and that's the key point in this whole argument", says the LLM. Around you go.
There's a better way. Instead of opening chatGPT and relying on answers built for the masses you need specific answers based on a much deeper context combined with rules and taste. Enter skills.
The quick version is you crack open a terminal, claude-code and codify the rules and process defining what good looks like.
You need to make your resume? Don't ask chatGPT. Build a skill defining what a good resume looks like. Pitch deck for VCs? Don't upload it and ask chatGPT - you'll go in circles. It's better to upload the deck and run a pre-defined skill using e.g., the best practices from YC, or top tier VCs/Angel groups. You'll get a concrete output to improve upon. And from that you can iterate in a straighter-line than you're likely to do in a generic LLM.
I publish mine here: https://www.ariakerstein.com/axestack