Thank you for featuring El Agente! ORCA is an incredible piece of software—powerful, reliable, and essential to making autonomous quantum chemistry possible.

Sounds great!

Those cases are likely to be semiautomatic at the current moment. Where human provide geometry and agent handle the execution and post processing. For example, here are xxx geometry, do calculation to verify the pathway…etc. as the system is bottle necked by geometry generation.

Transition state are much harder problem. My hypothesis is LLM can guess simple transition state but again building such structure is difficult without some flexible builder. We can potentially work on tool integrations like AutoDE to solve part of the problem but unsure how generalizable.

Absolutely. Those are identified limitations too. There are quite some limitations from Smiles to geometry especially those related to stereochemistry. We discussed this issue within the roadmap and limitations section. We will work on potentially a flexible molecule builder to address those issues

Link to preprint: arxiv.org/abs/2505.02484

@fchollet.bsky.social Hi François, we’re applying a dynamic hierarchical agent network to synthesize quantum chemistry workflows at runtime. This work is deeply inspired by your theory of deep learning–guided program synthesis. I believe our new cognitive architecture may be of interest to you.

Hi Dennis, would be interesting if you can come up with a prompt. Ideally not a huge system so we can check fast.