Blog.

The Altan Prompting Bible: Master AI Agent Communication

You know the feeling: you write a perfect prompt, hit enter, and the AI... misunderstands completely. Builds the wrong thing. Misses the point.

The problem isn't the AI. It's the prompt.

Agent-native development with Altan is different from traditional AI coding. You're not just prompting one model—you're orchestrating a team of specialized agents. Each agent has unique capabilities, and knowing how to communicate with them is the difference between mediocre results and production-ready software.

This is the Altan Prompting Bible.

The Golden Rules of Agent Prompting

Before we get into specific techniques, master these fundamental principles:

Rule 1: Be Specific, Not Vague

Bad: "Build a dashboard"

Good: "Build a dashboard with three sections: recent orders (table with columns: ID, customer, total, status), revenue chart (last 30 days), and top products (list with images and sales count)"

Agents execute what you describe. Specificity = Better results.

Rule 2: Describe the What, Not the How

Bad: "Create a React component with useState for form data and useEffect to fetch API data with axios"

Good: "Create a form to capture user info (name, email, phone). When submitted, save to the users database"

Let agents choose the implementation. They know best practices better than you think.

Rule 3: Context Is King

Bad: "Add a filter"

Good: "Add a filter to the products table that lets users filter by category (dropdown) and price range (slider)"

Agents don't have telepathy. Give them context about where, what, and why.

Rule 4: One Task, One Prompt

Bad: "Build a login page, add user auth, create a dashboard, and set up email notifications"

Good: Start with "Build a login page with email/password fields and a submit button"

Break complex projects into sequential prompts. Build incrementally.

Prompting Each Agent: The Complete Guide

Altan has four specialized agents. Each has different strengths and responds better to certain prompting styles.

Interface Agent: Building Beautiful UIs

What it does: Designs and builds user interfaces

Best practices:

  • Describe layouts: "Put the logo top-left, navigation top-right, main content centered"
  • Reference examples: "Make it look like Stripe's pricing page" or "Modern SaaS style with gradients"
  • Specify interactions: "When users click the button, show a loading spinner"
  • Be visual: Describe colors, spacing, typography

Example prompts:

"Create a hero section with a bold headline, subheading, two CTAs (primary and secondary), and a hero image on the right. Use a gradient background."

"Build a product card component: image at top, title and price below, add to cart button at bottom. On hover, show a subtle shadow."

"Make the dashboard more modern. Add more white space, use a cleaner font, and add subtle animations on hover."

Database Agent: Structuring Your Data

What it does: Designs database schema and manages data

Best practices:

  • List fields clearly: "Products table with: name, description, price, image URL, category, stock count"
  • Define relationships: "Each order belongs to a user. Each order has multiple order items"
  • Specify constraints: "Email must be unique. Price must be positive"
  • Think in entities: Users, products, orders, not technical tables

Example prompts:

"Create a users database with email, name, password hash, created date, and subscription tier (free/pro/enterprise)."

"Set up a booking system: properties (name, address, price per night), bookings (check-in, check-out, total price, guest ID), and guests (name, email, phone)."

"Add a reviews table linked to products. Each review has: rating (1-5), comment, user ID, and timestamp."

Flow Agent: Automating Everything

What it does: Creates automations, workflows, and integrations

Best practices:

  • Use trigger language: "When X happens, do Y"
  • Be sequential: "First, then, finally"
  • Specify timing: "Immediately" or "24 hours before" or "Every Monday at 9am"
  • Include conditions: "If the order total is over $100, send a thank you email"

Example prompts:

"When a new user signs up, send them a welcome email with their login details and a link to get started."

"Every day at 6am, generate a report of yesterday's sales and email it to admin@company.com."

"When an order is placed: 1) Send confirmation email to customer, 2) Create a notification for admin, 3) Update inventory counts."

Genesis Agent: Creating Custom Agents

What it does: Generates specialized agents to operate your software

Best practices:

  • Define the role: "Create a customer support agent"
  • List capabilities: "It should answer questions about orders, process refunds, and escalate complex issues"
  • Provide knowledge: "Give it access to the orders database and help docs"
  • Set boundaries: "Never issue refunds over $500 without approval"

Example prompts:

"Create a phone agent that answers customer calls, checks order status, and can schedule delivery changes. It should sound friendly and professional."

"Generate a data processing agent that reviews submitted forms, flags incomplete entries, and emails users to fix errors."

"Build a sales qualification agent: when someone fills out the contact form, the agent asks follow-up questions to understand needs, then books a demo with the right team member."

Advanced Prompting Patterns

Once you master the basics, use these advanced techniques:

The Iteration Pattern

Build incrementally:

  1. Start broad: "Build a product catalog"
  2. Add details: "Add filters for category and price"
  3. Refine: "Make the product cards look more modern"
  4. Polish: "Add loading states and error handling"

The Reference Pattern

Use real-world examples:

"Build a pricing page like Stripe's—three tiers with feature lists and prominent CTAs"

"Create an onboarding flow similar to Notion—step-by-step with progress indicators"

The Persona Pattern

Frame prompts from user perspective:

"As an admin, I want to see all pending orders in one view and approve/reject them with one click"

"As a customer, I want to track my order and get updates via email and SMS"

The Constraint Pattern

Specify what NOT to do:

"Build a settings page. Don't include password reset (we'll handle that separately)"

"Create an agent to handle refunds under $50. Don't allow refunds after 30 days"

Common Mistakes to Avoid

1. Too Vague

Bad: "Make it better"

Good: "Increase the heading font size, add more padding around sections, and use a lighter gray for secondary text"

2. Too Technical

Bad: "Add a useCallback hook to memoize the function"

Good: "Optimize the page so it doesn't re-render unnecessarily"

3. Overloading

Bad: "Build the entire e-commerce platform with products, cart, checkout, payments, and admin dashboard"

Good: "Start with the product listing page"

Practice Makes Perfect

The best way to master agent prompting? Build something.

Start with a simple project:

  • Personal portfolio website
  • To-do list app
  • Contact form with email notifications
  • Simple CRM for tracking leads

Pay attention to what works. When agents build exactly what you wanted, analyze your prompt. What made it successful? Reuse that pattern.

Start Prompting Better Today

The difference between good and great agent prompting is practice. Head to altan.ai, start a project, and apply these principles.

Remember: agents are powerful, but they need clear direction. Give them specificity, context, and incremental tasks—and they'll build things that surprise you.

Now go build something amazing.