I took and passed the GitHub Copilot GH-300 exam last week. That’s the third Microsoft exam I’ve passed this year! (The others were AZ-900 and AZ-204.) This exam was pretty easy. I did these two learning paths from Microsoft:
…and that was it, really. I worked through some of the example projects. And I made a point of trying to use GitHub Copilot for some stuff at work over the last couple of months.
Overall, I’m not that impressed with GitHub Copilot. It’s cool when it works, but when it gets confused, or gives me bad info, then I feel like I’ve wasted my time. Generally, for most stuff, I’ve found that searching Stack Overflow gets me better results than asking Copilot.
I haven’t found a single case yet where trying to get agent mode to do something even mildly complicated produces good results. I have, of course, already hit the issue where agent mode insists that it’s doing something, when in fact it is not doing anything. That’s frustrating. And once it’s gone down that path, you really can’t convince it that it’s lying/hallucinating/whatever.
Maybe I just haven’t learned all the ins and outs of prompting it yet. I’ll keep trying. I’m still not quite “drinking the Kool-Aid” on all of this AI stuff. There’s a lot of overblown hype out there. I do think there’s some usefulness to it, but it’s not as powerful as some people think it is.