Legacy: Generation-Era Workflows
These concepts reflect an earlier phase of ASA, focused on controlling how AI generates code. They are no longer part of the active ASA standard.
What This Archive Contains
Early versions of ASA included workflows designed to make AI code generation predictable and repeatable. These were built around the idea that structured enough instructions could make generation deterministic.
Deterministic regeneration
A process for regenerating parts of a codebase from structured specifications while preserving existing logic.
Marker-based code preservation
A convention for marking sections of code that should be preserved across regeneration cycles.
Contract-driven generation
Specifications that defined the intended structure of a feature before generation.
Spec-to-skeleton generation
A workflow that converted high-level specifications into code skeletons for AI tools to fill in.
Why These Concepts Are No Longer Active
Modern AI coding tools have changed how founders build apps. In practice: non-technical founders cannot write structured generation specs, AI tools are capable enough that controlling generation at this level creates more friction than value, and the real risk is not unpredictable generation — it is unsafe output.
The current ASA standard focuses on a different problem: not controlling how AI generates code, but verifying and enforcing the conditions under which AI-generated code remains safe.
How the Current Standard Addresses the Same Goals
| Legacy goal | Current emphasis |
|---|---|
| Deterministic regeneration | Consistent verification + bounded blast radius |
| Marker-based preservation | Slice boundaries + CI/CD guard |
| Contract-driven generation | Check-based verification (PASS/FAIL against defined criteria) |
| Spec-to-skeleton pipeline | ASA-aligned production foundation |
| Slice-level generation | Slice Architecture checks (verifies structure holds) |
Historical Note
ASA was originally called "Atomic Slice Architecture" and focused on making AI code generation structured and repeatable. The core insight about slices — self-contained vertical features with hard boundaries — remains valid and is the foundation of the current Slice Architecture layer. What changed is the framing: from "how to guide AI generation" to "how to keep AI-generated code safe over time."