alt Hacker NewsI personally have found success with an approach that's the inverse of how agents are being used generally.
I don't allow my agent to write any code. I ask it for guidance on algorithms, and to supply the domain knowledge that I might be missing. When using it for game dev for example, I ask it to explain in general terms how to apply noise algorithms for procedural generation, how to do UV mapping etc, but the actual implementation in my language of choice is all by hand.
Honestly, I think this is a sweet spot. The amount of time I save getting explanations of concepts that would otherwise get a bit of digging to get is huge, but I'm still entirely in control of my codebase.
Yep, this is the sweet spot. Though I still let it type code a lot - boilerplate stuff I’d be bored out of my mind typing. And I’ve found it has an extremely high success rate typing that code on top of its very easy for me to review that code. No friction at all. Granted this is often no larger than 100 lines or so (across various files).
If it takes you more than a few seconds or so to understand code an agent generated you’re going to make mistakes. You should know exactly what it’s going to produce before it produces it.