In my experience, it's been the opposite. I've personally had huge productivity gains. Code quality is decent, it typically aligns with whatever stage of the project I'm on (for instance, greenfield/PoC it might build out a bit of a monolith, but later when I switch out to modularizing and making the codebase scalable it does a fine job at that too).

I've used it for a ton of stuff now and so has my staff. Migrating between frameworks or upgrading to new patterns are both way faster and easier than I or any of my ICs could do it on our own. Hell, I even use LLMs to interact with JIRA at this point (write scripts to pull epics/tasks, create child tasks, etc then have the LLM process what it gets from JIRA and update accordingly, all interactively).

One thing I will say, I find the best value in the CLI based tools (claude code cli, gemini cli, openai codex). The IDE integrated tools just felt like a slightly nicer autocomplete/intellisense and would fall flat on any bigger requests.

Organization wide, I've noticed a few other anecdotes too. Juniors have no issue using an LLM, but since they don't know what they don't know it can get them running in circles on some stuff (Like no junior engineer, don't rebuild that whole module because you assumed it should be on a certain branch). Mid level engineers seem to echo the sentiment I see online most often, that they're great engineers and the LLMs are bad (I personally witnessed one change their tune on this once we forced them to use claude code instead of relying on IDE LLM integrations). And Senior+ engineers either fully embrace it and love it or they'll use it on occasion or for specific tasks (Maybe the latter are those super-ninja 10x engineers we've always talked about).

So my question, specifically for those of you that don't find LLMs useful and think they're worse for productivity: * What tools did you try? * What kind of work did you use them for? * How did you prompt the LLM? * How long did you give it a chance?