Skip to content

CORE Autonomy Ladder

The CORE Autonomy Ladder defines how far the system can safely act on its own. It is not a marketing concept — it is an engineering and governance model.

Each level represents a quantifiable capability that exists today or is planned, and every step is gated by constitutional controls.

The ladder ensures one principle:

CORE may become more autonomous, but never less governed.

Overview of the Ladder

Level Name Description Status
A0 Static System Manual development only Past
A1 Governed Generation AI generates code in crates, audited before integration Current
A2 Intent-Aware Agents Agents reason using Mind rules during generation Planned
A3 Governed Refactorer CORE proposes structural improvements through crates Planned
A4 Self-Evolving Constitution Agents propose constitutional changes, humans approve Vision

A0 — Static System (Historical Baseline)

What defines A0:

  • No autonomous generation
  • No agents
  • No crate lifecycle
  • No Knowledge Graph
  • Minimal or no governance

A0 existed only in the earliest experimentation phase — before the Mind–Body–Will architecture.

A1 — Governed Generation (Current Level)

A1 is where CORE stands today. It supports safe, governed AI-assisted development.

Capabilities:

  • Crate creation from natural-language intent
  • Autonomous coding and test generation inside crates
  • Validation pipeline (Black, Ruff, Pytest, syntax checks)
  • Full constitutional audits over generated changes
  • Accept/reject workflow
  • Introspection, knowledge sync, drift detection
  • Self-healing (IDs, docstrings, headers, style, import rules)

Guarantees:

  • Agents cannot modify files directly — only through crates
  • Mind audits everything before changes reach the codebase
  • No changes merge unless compliant with .intent/
  • Humans remain in full control of final integration

A1 forms the foundation for safe, incremental autonomy.

A2 — Intent-Aware Agents (Planned)

In A2, CORE becomes significantly more intelligent — not more dangerous.

Agents still cannot bypass the Mind or modify code directly. But they will:

Capabilities to be added:

  • Use .intent/ rules as active constraints, not passive filters
  • Read capability metadata from the Knowledge Graph
  • Reason about architectural boundaries during generation
  • Produce plans aligned with system domains
  • Make more accurate and consistent code suggestions

Guarantees:

  • Constitutional Auditor still has veto power
  • All changes still happen through crates
  • No autonomous modification outside governed pathways

A2 is about guided intelligence.

A3 — Governed Refactorer (Planned)

A3 introduces autonomous, but governed, refactoring.

Future capabilities:

  • Propose structural improvements via crates
  • Identify drift or anti-patterns and recommend fixes
  • Generate safe transforms for large or complex code areas
  • Use semantic understanding from the Knowledge Graph
  • Assist in organizing capabilities, domains, or symbols

Guarantees:

  • Refactors never bypass audits
  • All changes are explicit, contextual, and inspectable
  • Proposals remain small, bounded, reversible

A3 produces a system that can help maintain itself.

A4 — Self-Evolving Constitution (Vision)

A4 is CORE’s long-term North Star.

The idea is not autonomy without limits — it is governance with machine assistance.

Vision:

  • Agents identify when constitutional rules need adaptation
  • Propose amendments to .intent/
  • Humans review, sign, and approve
  • Canary validation ensures the new rules do not break CORE

Guarantees:

  • Mind always has human oversight
  • Agents cannot self-ratify any constitutional change
  • Evolution remains safe, traceable, and reversible

A4 is not the system governing itself — it is the system helping humans govern it.

Why the Autonomy Ladder Matters

  1. Prevents accidental overreach
  2. Ensures transparency in capability growth
  3. Gives developers a precise mental model
  4. Makes limitations explicit
  5. Aligns code, docs, and governance

CORE evolves only when the Mind allows it. Autonomy never exceeds governance.

Next Steps

Continue with:

Or return to the root: 📘 Home