It's the best AI code assistant I've seen so far, but it still just doesn't feel smart enough. I tried building a feature with it this morning but ejected after an hour or so. I love the way it incrementally makes changes and gives you full control, but the code it writes misses all kinds of semantics in the code (like distinguishing milliseconds vs seconds). Still hopeful I can figure out how to use it, maybe for obvious, large refactors.