The Agent Delivery Harness: An Opinionated Multi-Model Layer
An agent delivery harness turns human intent, experience, and judgment into accountable, evidence-bound decisions across one or more model runtimes.
14 min read
Designing the authority layer around agent-produced change.
Models propose. Systems preserve state, bind evidence to exact candidates, enforce policy, and keep consequential decisions with named humans.
Dakota Smith AI Systems Architect & Full-Stack Engineer daksmith.dev
Current analysis
Publication record
- 23 published analyses
- 84 field-guide pages
- 19 public system records
From position to proof.
One argument, carried through the publication: define the boundary, teach the machinery, inspect the implementation, and keep the unproven claims visible.
About the practicePosition
The Agent Delivery Harness
A falsifiable seven-invariant profile for accountable acceptance outside model sessions.
Read the conformance profileField guide
Delivery control above the agent loop
The machinery that separates runtime execution from evidence, policy, and authorization.
Study the control layerPublic system
delivery-harness
Durable work state, exact-candidate gates, content-addressed evidence, and bounded authority.
Inspect the implementationEvidence boundary
Local contract 7/7. Operational proof 0/4.
An author self-audit: process conformance is documented; live delivery and external acceptance remain unproven.
Review the stated limits
Practical knowledge for building agentic systems that ship.
A four-part field guide from work structure and portable capabilities through runtime control, delivery authority, and security.