I’m not saying that if we rub some AI agents on Obsidian, we’ve suddenly solved this longstanding problem
Programming languages are designed to let engineers specify exactly what should happen. “Higher-level” languages leave more details to the computer; “lower-level” languages take control of more details. Perhaps LLMs can enable an even higher-level language which specifies “just the important details”—an admittedly undecidable category—in English
The promise of LLM-generated tools is not just automation but co-design: functionality shaped in collaboration with the user so they understand the system’s logic as deeply as it understands their workflow. The tool becomes legible, not just useful. pkm
