AI LLMs will be a critical part of RTL development.
For documentation and logs. That's it, really. I'm never letting it touch consequential RTL in my life. Sorry for the bait.
But seriously, if you haven't tried giving an LLM access to the full documentation of whatever tool you're using, try it. Doubly so if you have an agent. I gave it Questasim documentation and the path to my latest run and I said "this compile crashed, tell me where and why". It was able to tell me the exact line where my coworker flubbed a patch he pushed to the compile script. His runs worked because he had existing libs in the simulation directory, but I was saved an hour of anguish grepping and chasing other errors my testbench was throwing. It's great at telling me all the flags and commands I need for Vivado, adding waves to a waveform window, and some limited success at diagnosing testbench bugs from a csv exported from said waveform. It's way better at grepping than I ever could be. Its tcl is good tool, so if you want a good set of signals to probe, it can give you a good starting position.