Agentic Instructions

Instead of specifying each action, you describe what you want to accomplish. The agent figures out how to do it.

​

Philosophy of Agentic Instructions

Agentic instructions operate on the principle of intent-based automation, where users describe what they want to accomplish rather than how to accomplish it. This approach provides:

• Natural Language Interface: Communicate goals in human terms
• Adaptive Behavior: Agents adjust to interface changes automatically
• Reduced Maintenance: High-level instructions remain valid despite UI changes
• Complex Reasoning: Agents can handle multi-step workflows autonomously

# Agentic approach: intent-based control
with VisionAgent() as agent:
    agent.act("Fill out the login form with username john.doe and submit it")
    agent.act("Navigate to the user profile settings")
    agent.act("Update the display preferences to dark mode")

​

The Agentic Approach in Practice

The agentic approach works by defining boundaries and rules rather than specific steps. The agent follows these guidelines to accomplish tasks autonomously. This method is particularly effective for:

• Complex Workflows: Multi-step processes with decision points
• Error Recovery: Handling unexpected states gracefully
• Cross-Platform Testing: Adapting to different device configurations

​

Improving Agent Instructions

One of the key advantages of agentic instructions is that you can ask the agent to improve the instructions themselves. Agents can:

• Identify gaps in the current instructions
• Suggest more robust error handling
• Recommend additional validation steps
• Optimize instruction clarity and effectiveness

This creates a self-improving system where the automation becomes more reliable over time.

​

Next Steps

• Review Single-Step Commands for comparison
• Explore Effective Agentic Instructions for best practices
• Learn about AI Models that power agentic reasoning
• See a practical Agentic Instructions Example for Android UI testing