For me, a big issue is that the performance of the AI tools varies enormously for different tasks. And it's not that predictable when it will fail, which does l

For me, a big issue is that the performance of the AI tools varies enormously for different tasks. And it's not that predictable when it will fail, which does lead to quite a bit of wasted time. And while having more experience prompting a particular tool is likely to help here, it's still frustrating. There is a bit of overlap for the stuff you use agents and the stuff that AI is good at. Like generating a bunch of boilerplate for a new thing from scratch. That makes the agent mode more convenient for me to interact with AI for the stuff it's useful in my case. But my experience with these tools is still quite limited.