It can write code faster than I, it can also quickly implement well known algorithms, but it is such a quagmire. You need to be very accurate in instructions and watch carefully to see if it is not suddenly having brilliant ideas and going around breaking everything else, regardless of what it was told to do. Like many people, I keep questioning if there is any actual productivity gain here, or if you just need to spend more time fixing the outrageous things AI produces.
The autopilot analogy landed for me. I've had the same experience where figuring out how to talk to the tool is half the battle. For anyone not locked into a single provider, OpenCode gives you that same workflow but with model choice baked in. I ended up documenting the whole learning curve from scratch: https://reading.sh/the-definitive-guide-to-opencode-from-first-install-to-production-workflows-aae1e95855fb
It can write code faster than I, it can also quickly implement well known algorithms, but it is such a quagmire. You need to be very accurate in instructions and watch carefully to see if it is not suddenly having brilliant ideas and going around breaking everything else, regardless of what it was told to do. Like many people, I keep questioning if there is any actual productivity gain here, or if you just need to spend more time fixing the outrageous things AI produces.