Show HN: Huzzah – a novel approach to coding with AI

August 2026

A caller experimental measurement to codification pinch AI

If you’re a package technologist for illustration me, the first fewer months of 2026 were incredible. Coding agents abruptly became bully capable that we nary longer needed to manually constitute code. But if you’re for illustration me, past sometime later you deed a wall. The honeymoon play ended, and the novelty wore off. No much dopamine hits.

It’s August, and I consciousness utterly fatigued. To beryllium honest, I’m sick to decease of penning longform English to picture each alteration I want to my codebase. However, I besides don’t want to spell backmost to penning each my codification manually. There was existent tedium successful that believe that I’d for illustration to debar for…well, the remainder of my life.

And yet, I consciousness that I need to person amended penetration and power complete what my codification is doing. I want to cognize that my output is precocious quality, reliable software. I want to consciousness bully astir myself arsenic a professional. So I’m trying to find a measurement to person my barroom and eat it too.

My problem pinch coding agents is that

1. There’s nary reliable grounds of quality intent. Prompts are discarded, and the codification whitethorn aliases whitethorn not person been generated by AI. We’ve mislaid the cardinal authority that expresses what the quality wants retired of the machine, and I deliberation it’s important to contend pinch that fact.
2. AI chats are imperative, step-by-step instructions that picture changes to the application, not the exertion itself. This intends instructions are often repeated, and frankincense devour tokens, galore times complete the people of development. This is inefficient.
3. Much of earthy connection exists for societal reasons, not informational. The mean condemnation is scarce successful existent information. Writing successful this manner, to a machine, is cumbersome.

---

To reside these problems, I’m building an experimental editor. I’m calling it Huzzah, and it poses an replacement paradigm for moving pinch LLMs.

With coding agents, prompts are (a) longform, (b) imperative, and (c) transient. With Huzzah, prompts are (a) pseudocode, (b) declarative, and (c) persistent.

It’s easier if I conscionable show you.

Your browser does not support embedded video. You can

watch the Huzzah objection directly

.

Comparing fizz buzz

Let’s return a very elemental illustration - opportunity you want to usage AI to create fizz buzz. We’ll do this doubly - erstwhile pinch coding agents and different pinch Huzzah.

With coding agents

You commencement a chat successful your instrumentality of choice, and type thing for illustration the following:

Create a usability that loops 100 times. If the number is divisible by 3, people “fizz”. If the number is divisible by 5, people “buzz”. If the number is divisible by some (like 15 for example), people “fizz buzz”.

If you request to make an edit, you’d nonstop a travel up connection to the chat:

Instead of looping 100 times, the usability should return a number input and the usability should loop that magnitude of times.

You repetition this process until you’re satisfied.

With Huzzah

You create a caller record called fizz_buzz.hz. In it, you constitute a pseudocode representation, nevertheless you like. This is really I’d do it, personally:

fizz_buzz() loop 100 modulo 3 ? "fizz" 5 ? "buzz" some ? "fizz buzz"

You prevention the file, and Huzzah automatically generates existent codification from it.

If you request to make an edit, simply update your file:

fizz_buzz(n) loop n modulo 3 ? "fizz" 5 ? "buzz" some ? "fizz buzz"

When you prevention the file, Huzzah captures the diff and uses it arsenic the punctual to the LLM. The affected root codification is frankincense regenerated.

Some different examples

To springiness you a amended consciousness for what this could look for illustration successful different scenarios, present are immoderate replacement examples.

1. Shopping cart

list cart list inventory mock_data = // see immoderate mock data init() inventory.fill(mock_data) add_item(id) cart.add(item by id) remove_item(id) cart.filter(item by id) checkout() return cart.sum(item by price) and format arsenic price

2. Todo List

Todo { id: int text: str completed: bool } add_todo(text) todos.add(text, completed = false) toggle_todo(id) todo = todos.get by id todo.completed = NOT .completed remove_todo(id) todos.filter by id

Benefits

You should beryllium capable to spot immoderate benefits already. Notice really overmuch much terse and readable the pseudocode is than the longform prompts? Here are immoderate more:

• Writing prompts this measurement engages your mind, because it feels overmuch much for illustration you’re designing the style of the code.
• You tin beryllium arsenic terse aliases arsenic verbose arsenic you like.
• The pseudocode acts arsenic developer archiving because a quality wrote it to definitive their intent.
• You could constitute a connection agnostic pseudocode and usage it arsenic the ground for aggregate connection aliases biology targets. Think analyzable algorithms, for illustration a CRDT.

Caveats

There are nary metallic bullets, of course. Some exceptions:

• It’s wholly imaginable that location are issues pinch this attack astatine scale.
• This is evidently much perfect for caller codebases than existing ones.
• If you deficiency domain expertise, earthy connection is astir apt the easier relationship method.
• Some things whitethorn beryllium much difficult to reliably express, for illustration cross-file dependencies.
• LSP-type features would not beryllium disposable (though this could plausibly beryllium generated).

Current state

Huzzah is actively being developed, and exists only successful an experimental authorities for now. You tin find the source codification and setup instructions here. Please springiness it a rotation and fto maine cognize what you think!

Cheers.