Yeah for me the accelerator has been using LLMs to build engines, not apps. Build something that outlasts the LLM, or could even be trivially used by a local LLM.

I'm under no illusion that we'll have cheap LLM access for much longer so I'm hedging my bets.