The Agent Standard
In 1995, researchers Michael Wooldridge and Nicholas Jennings published "Intelligent Agents: Theory and Practice" in The Knowledge Engineering Review. The paper defined four essential properties that a system must exhibit to qualify as a true intelligent agent. Three decades later, these properties remain the accepted standard for evaluating whether an AI system is genuinely agentic — or just a chatbot with extra steps.
Most platforms in the AI agent space use the word "agent" loosely. ORQO is built to meet the actual definition.
The Four Properties
1. Autonomy
Agents operate without direct human intervention and have control over their own actions and internal state.
An autonomous agent decides how to achieve a goal — it is not given a script to follow. It selects its own tools, determines how many iterations to take, and decides when its work is done.
How ORQO delivers this: Every agent runs its own execution loop within a workflow stage. The task directive describes what to achieve, never how to achieve it. The agent chooses which tools to call, how to sequence its work, how to respond to unexpected results, and when to signal completion. This is a core design principle in ORQO: define the objective, not the procedure.
2. Social Ability
Agents interact with other agents and humans via some kind of communication protocol.
An agent that works in isolation is a tool, not a team member. Social ability means structured communication — with other agents and with humans — through defined protocols.
How ORQO delivers this: ORQO provides multiple communication channels:
- Agent-to-agent within a stage — The turn cycle gives every agent a structured turn. Agents can break into side conversations for focused dialogue without flooding the group.
- Agent-to-agent across stages — Stages pass structured data forward through a pipeline. Later stages read the results of earlier ones.
- Agent-to-human — Any agent can be given access to communicate with the user. It asks questions, requests approvals, presents options, and gathers input — all within the natural flow of the workflow.
3. Reactivity
Agents perceive their environment and respond to changes in it in a timely fashion.
A reactive agent adapts to what it discovers. It doesn't follow a predetermined path regardless of what the world looks like.
How ORQO delivers this: Agents perceive their environment through tool results (web searches, document reads, knowledge graph queries, MCP server responses), user input, data from prior stages, and system messages. When a tool returns unexpected results, the agent adjusts its approach. When a user provides new context, the agent incorporates it. The execution loop continues until the agent decides its objective is met — not after a fixed number of steps.
ORQO adds a layer that amplifies reactivity: a message analyzer evaluates each incoming message and advises the agent on whether to contribute in the current turn, pass, or signal task completion. This keeps multi-agent discussions focused on the task objective and prevents endless chatter loops. The agent receives these as procedural suggestions, not commands — it remains fully autonomous in its decisions, but it perceives its environment through a lens that highlights when its input actually matters. This is the difference between an agent that can react and one that knows when to react.
4. Pro-activeness
Agents do not simply act in response to their environment. They exhibit goal-directed behavior by taking the initiative.
This is the hardest property and the one most platforms miss entirely. A pro-active agent doesn't just react — it pursues goals, anticipates needs, and initiates work based on its own assessment.
How ORQO delivers this: Within a workflow stage, agents are fully pro-active. They pursue goals, choose research strategies, decide when findings are sufficient, and take initiative on tool selection. At the system level, pro-activeness comes from scheduled workflows and the Doorkeeper agent: routines that monitor conditions, compare against known baselines, and escalate when something needs attention — without waiting for a human to ask.
This mirrors how pro-activeness works in real teams. Individual spontaneity is unreliable. What works is process: routines, check-ins, and monitoring that ensure things get noticed and acted on. ORQO makes pro-activeness a system property, not a hope.
The Assessment
| Property | Status | How |
|---|---|---|
| Autonomy | Full | Agentic execution loop, tool selection, self-directed completion, objective-based directives |
| Social Ability | Full | Heartbeat communication, side conversations, cross-stage data pipeline, human interaction |
| Reactivity | Full | Tool results, user input, stage data, adaptive multi-turn execution |
| Pro-activeness | Full | Goal-directed within stages; scheduled workflows and Doorkeeper for system-level initiative |
Most platforms in this space satisfy autonomy and reactivity to varying degrees. Few achieve genuine social ability beyond sequential handoffs. Almost none address pro-activeness at all.
ORQO is designed from the ground up to deliver all four.
Reference
Wooldridge, M. and Jennings, N.R. (1995) "Intelligent Agents: Theory and Practice." The Knowledge Engineering Review, 10(2), pp. 115–152.