Research: Work Is Distributed Across a Moving Determinism Boundary

Canonical principle: Principle 3: Work Is Distributed Across a Moving Determinism Boundary

Research Objective

Determine whether existing systems explicitly treat deterministic capability, procedural guidance, and AI reasoning as distinct operating mechanisms whose responsibilities can move over time based on execution evidence.

This is currently one of the most important differentiation questions for the Infoconex AI Flywheel.

Current Evidence

The current framework review shows several forms of movement between representations or mechanisms, but no reviewed framework has yet established the complete three-part, bidirectional model.

• Metis is the closest direct comparison. It explicitly decides when recurring procedural text should be promoted into validated executable code.
• GenericAgent crystallizes exploratory AI work into reusable SOPs, skills, scripts, and tools.
• MOSS moves learning into source-level implementation changes, but does not organize adaptation around a generalized choice among code, procedure, and reasoning.
• MASFly changes agent organization, collaboration procedures, and supervisory experience rather than explicitly moving responsibility into deterministic software.
• Ralph can improve code, tests, specifications, plans, and procedural instructions, but the responsibility boundary is implicit.
• Reflexion primarily keeps learning in linguistic reasoning memory.

The strongest prior-art pressure therefore comes from systems that convert recurring experience into more deterministic or reusable forms, especially Metis and GenericAgent.

Related Ideas and Historical Provenance to Trace

Research should identify earliest authoritative sources and establishment dates for:

• human-machine function allocation,
• levels of automation,
• progressive automation,
• skill compilation and proceduralization,
• program synthesis from demonstrations or traces,
• partial evaluation and specialization,
• adaptive workflow optimization,
• self-adaptive software that reallocates responsibility,
• transitions from learned or interpreted behavior into deterministic implementation,
• and mechanisms for reversing automation when deterministic behavior becomes brittle.

Open Research Questions

1. Has another methodology explicitly modeled code, procedure, and AI reasoning as three distinct but cooperating responsibility mechanisms?
2. Has responsibility allocation itself been treated as something the system should continuously optimize?
3. Is there a prior formal model equivalent to the term Moving Determinism Boundary?
4. Do existing systems support movement in both directions, including de-mechanizing brittle code back into procedure or reasoning?
5. What evidence should justify moving a responsibility from reasoning to procedure or from procedure to code?
6. What evidence should justify moving responsibility in the opposite direction?
7. Is recurrence enough to justify mechanization, or must stability, validation cost, consequence, and environmental variability also be considered?
8. How should governance limit movement across the determinism boundary when the change affects high-consequence behavior?
9. Does procedural guidance deserve treatment as a distinct mechanism, or can prior-art systems reasonably treat it as one form of text memory?

Evidence Still Needed

• Primary literature on function allocation and adaptive automation.
• Systems with explicit bidirectional movement between flexible and deterministic execution.
• Formal criteria for deciding when learned behavior should become code.
• Evidence of systems that reverse deterministic automation after brittleness is observed.
• Research establishing whether procedural knowledge is treated independently from general memory or prompting.
• Historical establishment dates for the closest related concepts.

Current Research Position

The movement of repeated behavior into reusable procedures or executable code is clearly established prior art.

The current differentiation hypothesis is narrower: the AI Flywheel explicitly optimizes where responsibility resides among deterministic capability, procedural guidance, and AI reasoning, allows evidence to move responsibility in more than one direction, and subjects that movement to human-defined governance.

Metis remains the most important framework comparison for testing this claim.