Dear Arnoud,

thank you for developing this! It is an amazing project.

I agree with your observations about agenting AI use. I can now have projects implemented that I never got around to. I am a bit concerned about a kind of "cognitive debt" - I hand over much of the ground work to programming agents, When teaching the Wolfram Language I notice that some students learn much faster, but many do not understand the underlying logic of the language anymore, as they often have an LLM generate the code but do not spend enough time to fully understand it.

My feeling was that LLMs often struggled a bit with the Wolfram Language and hallucinated functions that don't exist, but recently these problems do hardly show up in my workflow any more, especially if I give them access to wolframscript in the command line.

It sometimes feels that I increasingly have LLMs use the Wolfram Language to enhance their capabilities instead of using it myself. In one of my recent workflows I managed to give several agents access to wolframscript, and lean4 via the command line. Together these technologies managed to prove new theorems and develop new conjectures.

Sorry for rambling. Thanks a lot for posting - I am sure I will make good use of this paclet.

Thanks, Marco