Writing effective Agent prompts

The quality of what Agent builds depends heavily on how you describe it. Here are the patterns that consistently produce the best results.

Be specific about the stack

Too vague:

"Build me a web app"

Better:

"Build a React frontend with a Node.js/Express backend and a PostgreSQL database"

Describe the user experience

Agent thinks in terms of user flows. Describing what a user can *do* is more effective than describing what the code should *be*.

Technical description (harder for Agent):

"Create a JWT auth system with bcrypt password hashing"

User-centric description (better):

"Users can sign up with email and password. Passwords are stored securely. After logging in, users stay signed in for 7 days."

Include data models

Tell Agent what data you're working with:

"Build a project management app. Projects have a name, description, and status (active/archived). Each project has Tasks with a title, assignee, due date, and priority (low/medium/high)."

Reference existing patterns

If your project already has code, Agent will follow existing patterns. You can also be explicit:

"Follow the same error handling pattern as the existing /api/users endpoint"

Iterating

Start broad, then refine:

First prompt: get the basic structure working
Second prompt: add authentication
Third prompt: improve the UI
Fourth prompt: add the specific feature you care about most