Ask HN: What Are You Working On? (December 2025)

https://github.com/rabfulton/ChatGTK

This is actually pretty much as done as it's going to be (could use some nicer UI feedback, i.e. how you actually use the app) - it is actually just a demo for an effort I undertook to mod Datastar to support nested web components. I am writing it up as we speak!

Instructions: you have to answer three questions; each one will auto-submit once your response goes over 100 characters; the answer to the third question is your "post". It's a proof of concept of a friction intervention for social media to encourage slow thinking before posting (and hopefully reframing negative experiences in the mind, it's kind of dual purpose).

The idea is to treat LLMs as constrained components inside explicitly defined workflows: strict input/output schemas, validated DAGs, clear failure modes, and replayable execution. Most “AI unreliability” I’ve seen isn’t model-related — it comes from ambiguous structure and hidden assumptions around the model.

We’re exploring this through a project called FACET, focused on making AI behavior testable, debuggable, and reproducible in the same way we expect from other parts of a system.

Still early, but the goal is simple: less magic, more contracts.

iOS/JS/Android Libraries for turning traditional inputs/forms into audio conversations.

Have everything working and am happy with how the underlying data structures came out.

It is interesting because where I ultimately landed there is just an additional "state". So there is a UI state (Browser managed DOM), your app logic state (consumers manage), and a conversation state. The library just exposes events so you can keep the conversation state in sync with your other states. I started out visualizing a direct like: as a conversation flow, but you could configure it to navigate pages, click elements, or whatever through audio conversation.