A few weeks ago, I received an email from a GitHub employee working on Copilot. They wanted to talk to users about their paste experience. I don’t generally participate to surveys (out of laziness and “I don’t have time for this”-ness), but because I like GitHub and had a somewhat disappointing experience with Copilot when I tried it earlier this spring, I decided to share my honest feedback. To my surprise, they got back to me to schedule a call.

I would define myself as an AI sceptic, in the sense that I see some interesting uses but I’m still waiting for my mind to be blown, my workflow to be improved and my voice assistant to be useful for more than setting timers. That’s before I even start thinking about the ethics questions and the environmental impact that come with all that.

The call lasted a little bit more than 30 minutes and the person on the other side was very friendly (and they had 2 very cute dogs, another bonus of allowing people to work from home). I was able to talk about all the frustrations that caused me to abandon Copilot, which I will share here as well. Note, I mostly used Copilot on my Comics Outmash project, which I already talked about here.

• Comics Outmash uses Kotlin and its semi-official DSL to write HTML. Every time I used a suggestion from Copilot, the code would not compile straight away, I had to go back to add the missing imports, which caused just enough friction to annoy me.
• I used Exposed to connect to my Postgres database and it suggested methods that were deprecated more than once.
• I tried asking questions in the Copilot Chat, specifically about Supabase that I was struggling to integrate into my pull request workflow. It suggested to use sucommands on the supabase command line tool that simply did not exist (I’m pretty sure they were commands to run migrations on a ruby on rails project).

I understand that those may seem like minor issues, but to be fair, I already had one foot out the door when I started this experiment. I have since then tried new models on ChatGPT, Claude and the Cursor editor that everybody talks about and again, while impressive with some aspects, I remain mostly unphased. Like Molly White puts it, it feels kind of useful, but it’s hard to know if it’s really worth it.

I really appreciated this meeting, because it allowed me to take a step back and summarize what I think of all that (picture me waving my hands in the air as you read it please) and share what the main blockers were for me (again, ethics and the environmental impact). As a bonus, they gave me a 75USD gift card for my time and I was able to put that toward buying new toilets for my home, so that’s cool.