Matt Stokes · Working Paper
Constraint, Capacity, and Self-Maintaining Agency
A working paper on how rules for finite memory shape later action, and what a system must prove before memory governance can justify more autonomy.
Finite memory forces a system to decide what persists. This working paper examines how that selection starts to govern future behavior. It extends the machine-side argument behind Memory Is the Constraint.
The paper draws on research into agent memory and supervisory control conducted by the named authors. Most current agents still depend on an external harness to manage continuity through progress files, retrieval systems, or application policy. The paper asks what would change if a system began to carry and protect state because its later operation depended on it.
Self-maintenance names that narrower capacity. The term marks a testable change in how state persists and shapes future action. It does not claim consciousness.
What a product team would need
Once memory affects future behavior, it becomes part of permission and security design. People need a visible record of what was kept, where it came from, which actions may use it, and how to correct or remove it.
A team granting more authority to an agent needs evidence that the agent's carried state makes its behavior more predictable. That link remains unproven. Until it survives controlled testing, independent supervision stays outside the system and consequential actions remain bounded and reversible.
The PDF describes the competing explanations and the tests that could disprove the model.
Read the full working paper (PDF, 14 pages)
Related reading
Memory Is the Constraint turns the memory argument into a product proposal. The Future of the Interface applies the authority model to adaptive and generated interfaces.