AI-Assisted Development – One Year with GitHub Copilot

Note: This article was originally published on June 1, 2024. Some information may be outdated.

AI pair-programming tools are no longer hype. After using GitHub Copilot daily for over a year, I can say it’s changed how I code - and not just in small ways.

What Copilot Helps With

Here’s where Copilot saves me time every day:

• Boilerplate - It writes out useEffect, useState, API call scaffolding, even basic CSS and Tailwind classes.
• Unknown syntax - If I vaguely remember how to write a Prisma query or a new regex pattern, it fills the gap.
• Test cases - I write a function, and it often autocompletes good test scenarios using Vitest or Jest.
• String manipulation or parsing - These are boring. Copilot makes them fast.
• Markdown, README, docs - I don’t love writing these, but Copilot makes suggestions to speed it up.

What Copilot Gets Wrong

Copilot isn’t magic. It guesses.

• It sometimes makes things up.
• You still need to verify logic and types.
• It might pull in patterns that are outdated or not idiomatic.
• It doesn’t understand the big picture of your code - only what’s around your cursor.

How to Use It Well

• Be active - Treat Copilot like a junior pair. Let it start, but you edit.
• Edit heavily - Use suggestions as a base. Don’t trust without reading.
• Avoid for sensitive or license-restricted code - Generated output might not be safe to copy blindly.

Overall Impact

After a year, I’m more productive. I write more. I finish side projects faster. I explore unfamiliar libraries without getting blocked.

The key? I use it as a boost, not a crutch.

Copilot won’t replace you. But it can help you code more, try more, and learn faster - if you drive the process.

---

Originally published at letanure.dev