With recent AI tool explorations - with Claude Code, Figma make etc. - I realized something quite banal. Simple and plain prompting, yet still precise, seems to outplay hyper-optimized pre-prompted approaches.
I have tried to explain my intent and let ChatGPT create "the most optimal prompt." For creative work, such as image generation using Midjourney or Freepik spaces, this approach worked really well. Without taking notes or counting stats, anecdotally, I'd say it's better than just trying it with plain language.
For my more product and software related experiments however, I've find it to be counterproductive. I have tried the same "create the optimal prompt" approach for one experiment with different tools. Claude code, Antigravity + Gemini 3, Framer Workshop and Figma make.
All outcomes were rather mediocre, whereas Framer workshop and Figma make came up with the best results. In this case, "best" might be very subjective. But overall, they understood the idea best, proposed best layouts and didn't try to engineer.
Claude and Antigravity in contrast created small monsters for a simple form / chat idea with some logic. It did work, but it would've needed many iterations to get to a satisfying point. It felt like doing way too much and then having to trim off the fat.
I decided to start over again in Claude code. Same version, but a simplified procing logic - I've used a spec for instructions I previously created with the help of Notion Ai, since it had access to our studio related details - and provided claude with an instructions.md.
With that, all I started with was:
"build a super minimalistic chat bot like interface for a proposal estimator. fetch all details from instructions.md. the chat window should look like a modern, minimal, high quality chat interface, similar to chatgpt."
The initial version was already closer to what I've imagined. Far simpler, better ui, nailed the logic and flow.
Why I am writing this has actually less to do with AI and prompting, but with a finding for myself. Overthinking stuff, in maaany cases and situations will not lead to better a better outcome.
It sounds obvious and banal, but as someone who would call myself an introvert with tendencies of a perfectionist, this is something I have to remind myself quite regularly. Which, luckily, has become much better over the last years.