June 2026
Designing Agent Behavior Before Runtime
The missing discipline between institutional intent and agent implementation
Abstract
This paper establishes pre-runtime agent behavior design as a foundational discipline for consequential autonomous systems. It addresses a blind spot in contemporary AI governance: the missing design layer between institutional policy and agent implementation.
While monitoring, evaluation, and runtime controls are necessary, they do not establish what an agent was deliberately intended and permitted to do. This paper introduces the Agent Behavior Specification (ABS) as the authoritative, versioned record of deliberate behavioral intent in workflow-specific contexts.
The discipline encompasses seven core components: Workflow Objective, Consequential Decisions, Behavioral Directives, Authority Boundaries, Escalation Logic, Refusal Conditions, and Accountability Structure. Together, these establish how institutions should shape autonomous behavior before operational deployment.
This work provides both theoretical grounding and practical methods, with applications across financial services, legal, healthcare, supply chain, compliance, and other domains where agents make consequential decisions.
What's in this paper
The Core Problem
Why institutional intent cannot be discovered in production and why this is fundamentally a design problem, not a safety or compliance problem.
The Three-Phase Model
How consequential disciplines separate preparation, execution, and review—and why agentic AI collapses these phases.
The Agent Behavior Specification
The seven components that define deliberately authored agent behavior and what each component means in practice.
Bidirectional Traceability
How to ensure behavioral directives map forward to implementation and implementation choices trace backward to directives.
Cross-Functional Collaboration
How governance, product, design, engineering, operations, domain experts, legal, and risk work together to design behavior.
Practical Examples
Worked examples showing how pre-runtime design applies in financial services, legal, healthcare, and operations.
How to read this paper
New to the discipline
Start with the Core Problem section to understand why pre-runtime behavior design is necessary.
Implementers
Focus on the Agent Behavior Specification framework, templates, and practical examples.
Governance and risk
See how pre-runtime design differs from model governance, compliance frameworks, and runtime controls.
Researchers
Read the research directions section for open questions and opportunities for future work.
Citation
Nourkeyhani, A. (2026). Designing Agent Behavior Before Runtime. Anchored Agency. Version 1.0.
Download the complete paper