If you’re feeling overwhelmed by AI, you’re not alone. The author argues that less is more: the key to productivity is mastering the harness (GitHub Copilot) rather than chasing gimmicks. This post presents a simple 8-step workflow using existing Copilot features.
1. Pick a tool, any tool
GitHub Copilot is available in CLI, VS Code, Visual Studio, JetBrains, and the new Copilot app. The core workflow is consistent across tools. Beginners are advised to start with the CLI for a direct, text-based experience.
2. Turn on YOLO mode
YOLO mode (or “Allow All”) lets the agent execute commands without asking permission. This autonomy boosts productivity. For safety, run agents in sandboxes like GitHub Codespaces or dev containers.
3. Start with a prototype
AI makes prototyping easy. For example, ask for 20 mocks of a date picker web component in an HTML file. Prototyping uncovers nuances early, saving time and tokens. Use a medium-sized model like GPT 5.6 Terra or Claude Sonnet on medium reasoning, and stick with it for prompt caching benefits.
4. Plan methodically
Switch to plan mode and describe your goal. The agent will ask edge-case questions. You can install the “grill-me” skill for more aggressive questioning. Engage deeply with the model; your expertise guides the plan.
5. Implement with Autopilot
Autopilot is a built-in loop that ensures the model completes every item in the plan. It automatically uses subagents (Explore for reading files, General Purpose for complex tasks) and multimodel workflows without extra configuration.
6. Human review and iteration
Review the AI’s output and iterate. Insist on quality; don’t settle for “good enough.” Provide conversational feedback. For design, use a CSS framework like Postrboard. The author shows how to refine a date picker by fixing animations, contrast, and navigation.
7. Rubber duck the result
Request a Rubber Duck review from a different AI model family (e.g., Sonnet after using GPT). This catches blind spots. You can combine Rubber Duck with Autopilot for iterative improvement until diminishing returns.
8. Profit
Stage and commit, or move to the next feature. Start a new chat session for unrelated tasks. The author emphasizes that simplicity helps multitasking and context management. While advanced features like MCP servers and custom agents exist, mastering the basic harness is sufficient for most.
Images
