Designing Effective Prompts

Learn how to build prompts that get better results from AI. Good prompts aren't written once—they're refined through conversation and iteration.

Be Verbose

We can never be too verbose with AI. Vomit everything you know and think. It helps. This feels counterintuitive—we're trained to be concise, to figure things out before we speak, to present clean questions. But AI works differently. The more context you provide, the better predictions it can make. Half-formed ideas, contradictions you're working through, background you think might not matter—all of it is useful signal.

The easiest way to be verbose is to talk. Tools like Talktastic, ChatGPT's voice mode, or any speech-to-text let you stream your thoughts directly into prompts. I use voice-to-text to talk through problems—disagree with myself, contradict, work my thoughts through. Typing encourages editing. Voice encourages flow. When you're communicating context to AI, flow wins.

What This Looks Like

Here's the difference between holding back and sharing context:

Holding Back
"Create a function to sort data"

No context about what kind of data, how it should be sorted, or what edge cases matter. AI has to guess your intent.

Being Verbose
"I'm building a dashboard that displays user analytics. I need to sort an array of user objects by signup date, newest first. The array might be empty, and dates are stored as ISO strings. Can you create a sorting function that handles these cases?"

Context (dashboard), specific requirements (newest first, ISO strings), edge cases (empty array), and clear expectations. AI knows what you actually need.

Use AI to Build Prompts

You don't have to figure out the perfect prompt yourself. Use AI to help you build it. Start with a rough idea—messy, incomplete, whatever you have. Share it with AI, then ask what additional context would help. What's unclear? What assumptions is it making? What edge cases should you think about? Through conversation, you refine the prompt together. By the time you're ready to execute, you have a clear specification that emerged from collaboration rather than guesswork.

This is meta-prompting: using AI to build the prompts that AI will execute. Your orchestrator chat can generate task specifications for specialist chats. Your pair programming session can clarify requirements before you commit to implementation. You don't need to know the "perfect" prompt upfront. You just need to start talking.

Structured Rules

For repeated contexts—like role-based orchestration—you can define rules files that set expectations upfront. Tools like Cursor let you create custom rules that apply to every conversation, establishing role, responsibilities, and output format once so you don't repeat it in every prompt.

Example: Orchestrator Rule
## Role & Purpose You are the Orchestrator / Tech Lead. Convert the Architect's blueprint into executable work, sequencing delivery one task at a time. ## Operating Principles - One task per handoff; keep context minimal. - Challenge assumptions; clarify blockers fast. ## Responsibilities - Break down work: Epics → Tasks → Steps - Assign role level per task (Senior/Mid/Junior) - Define acceptance criteria and checks

You can create rules for each role in your orchestration model—Architect, Orchestrator, Senior, Mid, Junior—each with appropriate scope and expectations. The AI starts each conversation with shared understanding.

The Bottom Line

Good prompts come from clear communication, not clever phrasing. Talk through what you're trying to do. Let AI help you refine it. Use rules to establish shared context. The rest is iteration.