Your Prompts Are Terrible Because You're Typing Them

The model isn't the problem. You are. Specifically — the way you're talking to it. I've been building AI systems for two years and I still catch myself making these mistakes. Here's what actually changed my output.

i'm Mohd Mursaleen, an AI engineer based in Bengaluru. this isn't a framework. not a listicle either. just three changes that gave me immediate, measurable results.

Stop Typing. Use Your Voice.

i switched to voice dictation around three months back. my output didn't just get "better". it jumped categories. less back-and-forth. sharper first drafts. fewer generic answers.

reason is simple: typing makes you compress. your brain is moving fast, your fingers are not. so you skip context, drop nuance, and send the model a skeleton.

when you speak, you explain properly. same way you'd explain a problem to a friend. natural. complete. less self-editing in the middle.

just try once with your phone/laptop dictation. you'll feel the difference before you even hit send.

if you're still typing every prompt, that's probably your first miss.

The Agent Is Blank. Give It Everything.

before you write anything, remember this: the agent knows nothing about you. not your project. not your codebase. not what you tried yesterday. not even what "good output" means in your context.

every gap you leave gets filled with a guess. and that guess is the internet average — not your reality.

so give it the full context: your background, your goal, your constraints, what you've already tested, what done looks like.

treat it like a very smart new hire on day one. high IQ, zero context. the more you load upfront, the better it performs.

that leads to the most important prompting idea i've seen.

Build a World, Not a Prompt

This section was inspired by Varun Mayya's video — genuinely one of the best 10 minutes on AI prompting I've watched.

simple example. ask someone to draw a room.

if you say "draw a nice room", they'll guess what "nice" means. could be anything.

but if you say "draw a small room with a wooden desk, one window facing west, afternoon sunlight coming through, stacks of books everywhere, a half-drunk cup of tea sitting on the corner", now they can actually see what you see. no guesswork.

that's world-building. and that's the gap between prompts that work and prompts that don't.

Varun used Frank Herbert's Dune to explain this, and honestly it's one of the best analogies i've heard. Herbert didn't just write "big scary sandworm." he built the whole world around it first: spice economics, the religion, the sounds before danger, the fear on a Fremen's face when the ground starts shaking.

by the time the worm appears, your brain already built it.

prompts work the same way. don't just ask for output. build the world around the output.

when the model gets your background, constraints, tone, history, and goal, it stops guessing. the right answer becomes the most obvious answer.

here's the same task with two prompts: one vague, one world-built. same model. same task.

difference isn't the model. it's the context lore you gave it.

---

three habits: voice first, blank-slate awareness, world-building.

none of these need money. none need a better model. none need a paid tool. just a better way of thinking before you speak (or type).

if you want to see world-building in a real system, check my five-agent orchestration platform that shipped to 200 users on launch day. every agent there had a world-built system prompt.

---

Written by Mohd Mursaleen — AI engineer, Bengaluru. geekymd.me