Framework Definition

Persistence vs. Authority

Coined by Richard Ewing, AI Economist

Share:

Definition

Persistence vs. Authority is the foundational AI systems governance principle distinguishing execution duration from state-altering permission scope. Persistence measures how long an AI agent can execute unattended across background servers and workspace applications (such as Gemini Spark); Authority measures what records, databases, financial ledgers, and external communications the software is authorized to modify independently. Conflating persistence with authority allows flawed assumptions to silently spread across connected systems, creating compounding state drift and forensic recovery nightmares. Operating safely requires narrow read triggers, deterministic write allowlists, and unified rollback mechanics.

Why It Matters

Extending agent runtime without strict write boundaries creates silent distributed failures. While terminal-interactive tools (Claude Code) contain blast radius through human presence, unattended background agents require mechanical write allowlists and deterministic rollbacks.

How to Calculate

  1. 1Audit persistent service account permissions and revoke unconstrained write tokens
  2. 2Separate background read triggers from state-mutating execution endpoints
  3. 3Measure state drift risk thresholds across multi-turn autonomous agent loops
  4. 4Deploy cryptographic state-hashing and action admissibility verification

Related Articles

Deep Dive on the Blog

Explore the latest analysis and practical applications of this framework on the engineering economics blog.

Search the Blog Archive →

Citation

To cite this definition:

Ewing, R. (2026). "Persistence vs. Authority." richardewing.io.
https://www.richardewing.io/articles/frameworks/persistence-vs-authority

⚡

Want to apply this to your organization with Persistence vs. Authority?

Run a free diagnostic first. If the numbers concern you, book a session to build a remediation plan.

Richard Ewing: AI Economist & Capital Auditor