[jillesvangurp]

Llm to text to diagram is the killer workflow these days.

The key challenge is making these things presentable. Optimizing them for human editability is a secondary concern at this point. This is where a lot of these tools fall apart.

[justonceokay]

I use LLM to tell my children bedtime stories and to wish my wife goodnight. Saves me a lot of time and energy

[billyp-rva]

> Llm to text to diagram is the killer workflow these days.

It's useful, but I think only marginally. When using a diagrams-as-code tool, having an LLM write it for you gets you up and running faster, but long term you have more control (and no hallucinations) just writing the code yourself. It'll probably be fewer keystrokes in the end. Doubly so if your diagrams-as-code tool has decent autocomplete.

[thejazzman]

i think d2 looks sick , and i'm not op, but i do love just describing my system and telling the LLM to draw it. so having it "draw" d2 seems like a great new step?

when you want to make changes, you update your docs/design and repeat. think system design interview but you skip drawing it yourself at all and then sanity check your own work by interpreting the resulting graph etc