6 comments

[ 2.8 ms ] story [ 23.3 ms ] thread
Coffee is a new OSS framework which lets you build and iterate on your frontend code – right from your own IDE.

It's currently focused on React, and the team at Coframe behind the project has experimented with a few distinct DX flows for how a frontend-focused Copilot should work.

The best part of the project is there's no real dependencies – just run one docker command, add `<Coffee>instructions</Coffee>` somewhere in your project, and hit save to see the magic happen.

It's currently pretty slow and has a bunch of rough edges imho, but I'm excited to see more experiments around generative UI in this space – and in particular, how we can build AI-based solutions which work well alongside a developer's existing workflow instead of taking them out of it – or worse, trying to replace them?

We have achieved caffeination internally
the hottest programming language is really natural language
This is a giant leap forward
Some background on why this DX: - you can pass real props and see how the component works. - you work in the same dev env, as you already do (IDE/Editor) - It leads to splitting the code into smaller blocks (which is good for LLM and humans).
what were the other DX flows you considered, and in your experimentation, what were the tradeoffs you found?