Refusal
The tool does not grant permission.
Only process audio you own, created, licensed, or otherwise have permission to use. Technical capability is not rights clearance.
Field receipt 001 / public-safe
Software built like a conversation—then made answerable to reality.
Natural Language Software Development is a human using language, taste, iteration, testing, and judgment to build real systems with AI agents. The agent can take initiative. The human remains responsible for what the work means and what leaves the room.
In plain language
The human does not disappear when the agent starts moving.
Chaz names the need and supplies the lived context. An agent can inspect, propose, code, document, and test. Chaz can disagree. The work changes. Tests reveal what is actually true. The shared result is neither a wish nor a magic trick: it is a reviewed artifact.
One real build
Signal to Stems, in eight human-readable beats.
This is a compact public receipt, not a transcript. Private prompts and reusable internal process stay sheltered.
-
01
Human intention
Make limited resources less limiting.
Chaz wanted a musician to turn a reference into something studyable—without pretending software created the musician’s imagination.
-
02
Source material
Lived music practice became the brief.
Money, gear, studio time, band logistics, and the gap between an idea and a workable artifact supplied the actual problem.
-
03
Language
The build began with sentences, not a perfect specification.
The working direction named the desired experience, artist-control rules, rights boundaries, and proof state. The exact private prompts remain private.
-
04
Agent initiative
The agent handled structure and mechanical drag.
It helped shape a clean standalone release, organize the local workflow, document known limits, and turn the prototype into reviewable public proof.
-
05
Taste + disagreement
The first wording showed too much plumbing.
An early public description leaned on a backend source name. Chaz’s judgment was simple: describe what a person can do, not the machinery underneath.
-
06
Testing
The story had to survive contact with the tool.
The local launcher was checked from a cold start and an immediate restart. Both reached the app successfully; the public repository and ethics boundary were checked separately.
-
07
Correction
The interface returned to the human action.
Backend-facing language became “supported track link,” “study-ready stems,” and “Create stems.” The code still does technical work; the visitor no longer has to speak like the backend.
-
08
Shared result
A useful prototype, with its limits still attached.
Signal to Stems is public portfolio proof and local/invite-only software—not a rights-clearing service, not a replacement musician, and not a claim that every separation will sound perfect.
Where the method stays alive
A refusal and a capability limit belong in the receipt.
Capability limit
A browser cannot quietly finish the local filing for you.
It can offer a download, but it cannot silently place stems into arbitrary folders on another person’s computer. The workflow has to respect that boundary.
Two different modes
“Made with AI” can describe very different relationships.
Both modes require human responsibility. They should not be flattened into one story.
Generative AI art
The model creates visible or audible material.
A person prompts, selects, edits, contextualizes, and remains responsible for rights and use—but the model generates part of what the audience directly sees or hears.
Human-led art made easier by AI tools
The human work is the source; the tool removes friction.
Signal to Stems belongs here. It helps a musician study and organize sound. It does not decide what the song means, perform it, or become its author.
More hands. Same voice.IP shelter
Show the method without emptying the workshop into the street.
Public here
Intention, decision points, one correction, test shape, ethical limits, proof state, and the public artifact.
Still protected
Private prompts, unpublished source material, credentials, personal data, and reusable internal process detail.
Natural Language Software Development is not asking the machine to be the artist. It is one human using language to give the work more hands—then using taste, tests, and responsibility to keep the voice human.