What's this about?
A Show HN proposing that you direct a coding agent by writing structured pseudo-code in a persistent file rather than prose in a chat window — so the specification lives alongside the code and is versioned with it.
What exactly is the abstraction problem here?
There's a spectrum. Real code is unambiguous but you have to write all of it. Natural language is fast to write but ambiguous, and the record of what you asked for evaporates into a chat log. Pseudo-code sits between: structured enough to pin down control flow and intent, loose enough that you're not writing the implementation. The second half of the idea matters as much — putting it in a file means the description of what the software should do is a durable, reviewable, diffable artifact rather than scrollback. That's an old idea (literate programming, formal specs) meeting a new reason to want it.
What's the community saying?
- The framing that got the most agreement: engineers empowered by LLMs are collectively searching for the right abstraction level to operate at, and prose in a chat window is probably not it.
- The sharpest challenge: why not simply tell your existing harness "if I give you pseudo-code, spell out my intent, then write and test it"? You can use pseudo-code today without a new tool.
- A comparison worth following: someone had reached for semi-formal specification languages like Quint for the same reason — wanting something more structured than prose.
- The scaling question nobody answered: how do you express multi-file changes, imports and exports — or are these files disposable per change?
- A dark-mode contrast bug meant several people couldn't read the examples at all, which is its own comment on design.