This is a submission for the GitHub Copilot CLI Challenge
What I Built
As a junior frontend engineer who recently graduated from college, I found that the hardest decision I make every day isn't about code architecture—it's "What should I eat for lunch or dinner?"
I built the Restaurant Slot Picker, a fun and interactive web application designed to solve daily "choice paralysis" when it comes to food. It identifies your current location, finds nearby restaurants, and presents them in a slot machine interface. With a single click, it randomly selects three options for you. If you see something you like, clicking on the card takes you directly to Google Maps for navigation.
And the three options is independent selected, which means maybe you would get three same results, which means you are lucky and you must go to try the restaurant.
Demo
Here is the live application and the repository:
How it works
When you first visit the page, you will see this interface:
Next, click the "Spin the Slot Machine" button.
Wait a few seconds, and the app will reveal three dining options for you, like this:
Finally, pick the one that looks best to you. Click the card, and the app will navigate you directly to Google Maps.
My Experience with GitHub Copilot CLI
Before this challenge, I exclusively used GitHub Copilot Chat within VS Code. Although I knew GitHub had released the CLI feature, I had never tried it out. I really appreciate this challenge for giving me the push I needed to finally explore it.
The Learning Curve
At first, I used the Copilot CLI with a bit of nervousness because it has the power to execute operations on my project. Since I wasn't entirely familiar with CLI commands, I felt the need to verify every action it proposed. This "forced learning" actually became a benefit—I had to understand various command-line instructions and their meanings before approving them.
From "Micromanager" to "Boss"
Initially, progress was slow because I was cautious and didn't fully trust the tool. However, once I built that trust and started allowing it to execute commands autonomously, the development speed skyrocketed. It could build features automatically while I literally sat back and watched YouTube, occasionally checking the results or providing new specifications. It felt incredibly "chill"—like I was a boss simply chatting with a developer and delegating tasks.
Beyond Code: APIs and Deployment
I also discovered that GitHub Copilot CLI is capable of much more than just writing logic. It acted as a comprehensive guide for infrastructure tasks, such as applying for the Google Maps API and deploying the project to Vercel. Instead of leaving me to read through long documentation, it generated a clear, step-by-step guideline. By following the instructions it provided, I was able to smoothly complete these configuration missions without getting stuck.
Finding the Balance: Direct Agent vs. Strict Specs
I also realized that for complex features, the agent might misunderstand my intent. I researched how to better converse with AI agents and experimented with defining strict project specifications (Spec-driven development).
While this process is more rigorous, I found it can sometimes lead to "over-engineering"—the AI might write tests or functions that aren't needed yet, making maintenance difficult. In contrast, the direct Agent mode is faster and easier for smaller tasks. I learned that the key is knowing when to use which strategy.
Conclusion
I also tried the copilot delegate features, which allow the agent to run in the cloud and create PRs for review. It makes me feel like I have the capabilities of a senior engineer, even though I just graduated!
Ultimately, this tool is incredibly cool. The biggest takeaway for me is that communicating with AI is a skill in itself. If the AI doesn't meet my expectations, it usually means my description needs improvement. I can't wait to see what else I can build with it.
Top comments (0)