Show HN: Huzzah – a novel approach to coding with AI
Points and comments are a snapshot, not live.
Huzzah is an experimental editor using pseudocode as persistent, declarative prompts for LLMs.
Huzzah addresses three problems with coding agents: prompts lack a reliable record of human intent, are imperative step-by-step instructions rather than describing the application itself, and much natural language is socially redundant. The editor uses pseudocode which is declarative and persistent, automatically generating real code on save. Changes are made by editing the pseudocode, and diffs are sent to the LLM. Benefits include terseness, readability, and the pseudocode acting as developer documentation. Caveats include unknown scaling for large or existing codebases and difficulty with cross-file dependencies.
What commenters are saying
Commenters found the idea interesting and aligned with existing explorations into semi-formal specification languages. Several questioned whether the editor is needed when one could already prompt an agent with pseudocode; the author emphasized that the novelty is persisting both intent and source maps between pseudo and generated code. Another camp debated whether this is the right level of abstraction, with some preferring to test-driven generate code or use CLI tools rather than a custom web UI. A few pointed out potential issues with ambiguity and debugging when intent is not translated correctly.