Developers are tired of chasing every new AI tool, MCP, or prompt trick that claims to solve everything.
The reality is that less configuration makes more difference. Most productivity gains come from understanding the core interface, not installing custom agents or writing complex workflows.
This guide outlines a simple way to improve your workflow using the standard GitHub Copilot harness. It requires no special skills and avoids gimmicks.
Disclaimers
The term “harness” refers to GitHub Copilot. The goal here is simplicity, so treat Copilot as an agent harness for this discussion.
This does not mean you will never need custom skills or complex automation. Those become important as you scale and build team processes. I use a few of those tools in this post.
The key point is that you do not need those extras to be highly successful with AI.
There is a lot of low-quality output available. If you ask an agent to create a skill for any task, it will do so. Whether that skill works is another matter. You can easily publish such outputs to registries.
1. Pick a tool, any tool
Choosing a platform is straightforward. Options include the CLI, the GitHub Copilot app, VS Code, Visual Studio, and JetBrains.
These experiences are increasingly centralised around the same harness. Details vary by tool, but the core workflow remains consistent. Learn the harness once, and use it everywhere.
Learning the harness is key, and the best method is to stay close to the interface. For beginners, the GitHub Copilot CLI is recommended. It is a text-based terminal interface with minimal UI. You enter a prompt, the agent acts, and the interaction feels direct and satisfying.
This demonstration uses the GitHub Copilot app. The underlying harness is the same one found in the CLI, VS Code, and other supported environments.
2. Turn on YOLO mode
YOLO mode is also known as “Allow All.” This allows the agent to execute commands without asking for permission. This varies by tool, but often involves a `/allow-all` command in the chat.
Without this, the agent stops and waits for approval every time it needs to perform work. This reduces productivity and creates a poor user experience. Approving every step trains you to stop reading what you are approving.
Safety is still required. Bad things happen to good people. When using YOLO mode, do not run the agent on your local machine. This is especially true for work environments where data privacy is critical and mistakes are costly.
Several options exist for running agents in sandboxes. GitHub Codespaces or development containers are easy starting points.
3. Start with a prototype
AI allows you to prototype anything quickly. Historically, prototyping was a full project phase. Now, you can create one with a prompt.
Consider building a date picker web component. This seems straightforward but involves many details.
- How does the user navigate the component?
- What does the selected date look like?
- What does a selected range look like?
- How do users navigate between days, months, and years?
Start with a simple prototype and generate several variations. A typical request is:
Give me 20 mocks for a date picker web component. Put them all in an HTML file so I can compare.
The AI generated various layouts. One mock starts with the year view. This is useful because you want the user to zoom out to the year, then into the month, and finally to the day. These details often do not appear until they are seen.
Humans process sensory-rich models like images and layouts faster than dense text. Creating low-effort prototypes early makes complex concepts intuitive.
This applies to non-visual tasks as well.
If adding a new API endpoint, create a visual mockup first to understand requirements. For example:
Create a visual mockup of the API for this project. Add five options for how we could handle a new API endpoint that allows the user to download their analytics data.
