Same.

• i think of search as a summary of the first page of search results. It takes slightly longer to come back but might save you time evaluating. But much of the time you do need to click into original source
• ai writing unfortunately is valued at my company. I suppose it helps us engineers write more effective docs, but they don’t really add value technically, and they’re obviously ai. I’ve used this to translate technical docs into wording so management can say “look how we use ai”
• ai coding is better. I use it through my ide as effectively an extension of autocomplete: where the IDE can autocomplete function signatures, for example, ai can autocomplete multiple lines. It’s very effective in that scenario
• I’m just starting with more complex rulesets. I’ve gotten code reviews with good results, except for keeping me in the loop so it inevitably goes very wrong. I’ve really polished my git knowledge trying to unwind where someone trusts ai results without evaluation but the fails forward trying to get it to fix itself until they can’t find their way back. This past week I’ve been playing with a refactoring ruleset (copied from online). It’s finding some good opportunities and the verbal description of the fix is good, but I’ll need to tweak the rule set for the generated solution to be usable

The short version is it appears to be a useful tool, IFF you can spend the time to develop thorough rulesets, stables of mcp servers, and most importantly, the expertise that you could do it yourself