The intern use case is 50% of the value I get from LLM’s (I refuse to call them AI because they aren’t). The other 50% is as a better stackoverflow where I can look up syntax etc.
Anything that requires a deterministic outcome is not a good use case for them. Anyone who says different is selling snake oil.
One interesting kind of cog - I've seen examples of tools that claim to provide reliable workflows of AI's as web-scrapers, to take a URL and pull out a list of JSON objects from a listing page (containing a set of specified fields) , and instead of having to write parsing code and deal with changes to the website, the AI driven scraper tool will just keeping chugging along.
Caveat - I haven't actually used these myself, like you I am a heavy user of AI interns!
They are a massive force multiplier for me in coding and in doing investment research.
I use cursor AI for generating code changes from feature descriptions, and perplexity-AI to answer complicated questions using current data from the internet.
Claud and ChatGPT are very nice for language translation - I use one to translate and the other the reverse translate to verify the quality of the translation.
Anything that requires a deterministic outcome is not a good use case for them. Anyone who says different is selling snake oil.