Learning Timeline
Key Insights
The Problem with 'One-Shot' Prompting
Building a startup or complex application using AI via 'one-shot' prompting usually fails within the first 30 minutes. Use 'agentic design patterns' where agents interact with one another for more stable and reliable results.
The 'Harness Engineering' Concept
This is an OpenAI term for process design that allows agents to perform peer-to-peer QA. Instead of manually monitoring for minor errors, you can build a dedicated 'Design Agent' to automatically detect UI bugs.
Cost vs. Management
Building an automated QA Loop will consume more 'tokens'. Ensure your API budget is sufficient before implementing agent systems that monitor each other continuously.
Prompts
UI Alignment Fix Prompt
Target:
AI Coding Agent
The agent mentions pill is misaligned with the project mention pill. Please adjust the heights and alignment based on this screenshot to make them consistent.
Step by Step
How to Provide Visual Feedback to AI Agents
- Identify visual errors or UI inconsistencies generated by the AI Agent (e.g., mismatched button heights or 'pill' components).
- Take a screenshot of the specific area where the problem occurs.
- Click on the 'Paperclip' icon or the file attachment area in the AI Agent chat interface.
- Upload the screenshot you just took.
- Type a specific description of the issue in the message box to reference the screenshot.
- Press the 'Send' button or hit 'Enter' to initiate the repair process (QA Loop) by the AI Agent.
