The Running Order

The Running Order is the universal runtime standard that every Post inherits. Where The Groundwork defines what a specific agent does, The Running Order defines how every agent behaves — regardless of its function.

Think of it as the employee handbook for AI agents. A customer support agent and a data pipeline agent do completely different work, but both must: log their actions, respect scope boundaries, escalate on ambiguity, never take irreversible actions without approval, and degrade gracefully when upstream systems fail.

The Running Order exists because without a universal standard, each agent develops its own behavioral patterns based on whoever built it. One agent logs everything. Another logs nothing. One escalates aggressively. Another never escalates. The inconsistency makes the system impossible to audit and impossible to trust at scale.

The Running Order covers five domains: execution standards (how agents process tasks), communication standards (how agents report status and surface issues), failure handling (what happens when things go wrong), boundary enforcement (how scope limits are maintained at runtime), and audit requirements (what gets logged and who can review it).

Each domain specifies the default behavior that every Post inherits. Individual Posts can tighten these standards in their Groundwork — a financial agent might require stricter logging — but they cannot loosen them. The Running Order is a floor, not a ceiling.

This inheritance model means new agents start from a safe, governed baseline. The team does not need to re-specify basic behavioral rules for every deployment. They write The Groundwork for what is unique about this agent, knowing that the universal standards are already in place.

In software engineering, the closest parallel is a base class or interface that all services implement. In DevOps, it is a compliance baseline. In traditional management, it is the difference between a company policy manual (universal) and a job description (specific).

EOS has a similar concept in its Core Processes — documented ways of doing fundamental work. The difference is that EOS Core Processes are designed for human teams. The Running Order is designed for AI agents whose failure modes are fundamentally different: they do not get tired, but they do hallucinate. They do not get distracted, but they do drift.

The Running Order is also what makes auditing possible. When every agent follows the same logging, escalation, and boundary enforcement standards, you can audit any agent using the same process. Without it, auditing requires understanding each agent's unique behavioral patterns — which does not scale.