You’re staring at a cursor. The prompt is almost right. But maybe you should rephrase that first part. Actually, should you add more context? What if it misunderstands? Maybe start over.
Sound familiar?
Loading...
You’re Not Writing a Legal Contract
Every extra second you spend "perfecting" a prompt is a second you’re not getting an answer. And here’s the uncomfortable truth: your polished version probably isn’t meaningfully better than your rough draft.
The model doesn’t grade your grammar. It doesn’t judge run-on sentences. It’s not going to think less of you. It’s just trying to help.
The Compound Cost
Say you spend 8 seconds per prompt instead of 2. Doesn’t sound like much, right?
Twenty-one hours. That’s nearly three full work days of your life, every ye…
You’re staring at a cursor. The prompt is almost right. But maybe you should rephrase that first part. Actually, should you add more context? What if it misunderstands? Maybe start over.
Sound familiar?
Loading...
You’re Not Writing a Legal Contract
Every extra second you spend "perfecting" a prompt is a second you’re not getting an answer. And here’s the uncomfortable truth: your polished version probably isn’t meaningfully better than your rough draft.
The model doesn’t grade your grammar. It doesn’t judge run-on sentences. It’s not going to think less of you. It’s just trying to help.
The Compound Cost
Say you spend 8 seconds per prompt instead of 2. Doesn’t sound like much, right?
Twenty-one hours. That’s nearly three full work days of your life, every year, that you could get back by just pressing enter.
The Model Can Ask Questions
This is the part people forget: LLMs are good at clarifying. If your prompt is vague, they’ll ask. If you’re missing context, they’ll tell you. The whole point is that it’s a conversation, not a one-shot command.
Tools like Cursor and Claude Code have literally built this into the product. "Plan mode" exists because the developers know you don’t have the perfect prompt ready. You describe what you want, the model proposes a plan, you refine it together. It’s designed for iteration.
I’ve watched engineers spend 3 minutes crafting a "perfect" prompt when they could have:
- Sent a 5-second rough draft
- Gotten a clarifying question
- Added the missing context
- Had their answer
Total time: 30 seconds. Better outcome too, because the model guided them toward what they actually needed.
When Precision Matters
There are exceptions. System prompts that run thousands of times? Worth optimizing. API calls you’re paying for at scale? Measure twice.
But for the 99% of prompts that are just you and a chat window? Send the rough draft. Iterate. Move on.
Beyond Prompts
This isn’t just about LLMs. The same pattern shows up everywhere.
The email you’ve rewritten four times? Send it. The message to that person you’ve been meaning to reach out to? Send it. The side project you’ve been "planning" for months? Ship it. The shot you’re not sure you should take? Take it.
Perfectionism masquerades as quality control, but it’s usually just fear. The rough version gets you feedback. The polished version sits in drafts.
Try It
Next time you catch yourself rewriting, stop. Look at your rough draft. It’s probably fine.
Just press enter.