Tootfinder

While I'm talking about new ways to structure computer programs, what does it look like to _push_ data into an LLM? Instead of making it natural language fetching, with huge privilege to do stuff, can we structure things to, say, have one system supply some data context, and push it into an LLM to arrange processing, and that in turn pushes to other systems for action? It's not quite “code and data are separate", but I think the inversion might help mitigate a lot of the lethal trifecta. It also puts humans in the artisan-director seat, rather than the wannabe slave-master's seat, metaphorically.