# Agentic AI > **Machine-Evolution Nexus:** [[articles/Digital Darwinism and the Invisible World of Machine Evolution|Digital Darwinism and the Invisible World of Machine Evolution]] places this node within the history and governance of substrate-independent evolution. **Domain:** Artificial Intelligence / Autonomous Action **Doc Type:** Canonical Wiki Hub **Maturity:** Developed ## Definition **Agentic AI** is the family of artificial-intelligence systems organized to convert objectives into sequences of consequential action. Such systems can plan, choose tools, preserve or retrieve state, observe environmental results, revise their approach, and continue until they reach a stopping condition or require human judgment. The term describes an architecture, not a claim about consciousness. A system can be highly agentic in execution while possessing no demonstrated subjective experience, moral autonomy, or independent purpose. ## The Governing Distinction [[wiki/Agentic|Agentic]] names a degree or property: how far a system can carry an objective through action and feedback without step-by-step direction. An [[wiki/AI Agent|AI agent]] is a particular actor that exhibits that property. **Agentic AI** is the larger technical category containing agents, model-directed workflows, multi-agent systems, and the supporting runtimes that make their action possible. ## Core Architecture An agentic system is assembled from several separable layers: 1. **Objective and constraints** — the desired completed state, budget, deadlines, prohibitions, and success conditions. 2. **Reasoning and planning** — decomposition of the objective into actions, with the ability to revise the plan. 3. **Tools and effectors** — APIs, browsers, terminals, applications, messaging systems, payment rails, code execution, or physical devices. 4. **State and memory** — working context, durable preferences, prior decisions, task history, and provenance. 5. **Runtime and persistence** — an execution environment that can survive beyond a single response or application session. 6. **Observation and verification** — results returned from the environment so the system can distinguish an attempted action from a completed one. 7. **Identity, permissions, and approval** — a record of whose authority the agent exercises, what it may access, and which irreversible actions require consent. 8. **Audit and recovery** — logs, checkpoints, stopping conditions, exception handling, and a way for a person to inspect or take over. The language model is only one component. The shift from a chatbot to an agent occurs when language becomes connected to stateful execution and environmental feedback. ## Workflow, Agent, and Persistent Counterpart A [[wiki/Agentic Workflow|workflow]] follows predefined paths even when models perform some steps. An agent dynamically selects its own path. A [[wiki/Personal AI Agent|personal AI agent]] adds continuity around a particular person: memory, accounts, permissions, routines, and a continuing working relationship. Persistence changes the unit of interaction from a disposable chat to a maintained counterpart. The consumer-facing branch is therefore not defined by a friendlier chat window. It is defined by the movement from **answering a request** to **accepting responsibility for an objective**. ## Product and Conceptual Lineage - **[[wiki/Facebook M|Facebook M]] (2015–2018)** tested the intent-to-outcome abstraction inside Messenger. Human trainers completed much of the procedural middle, revealing both demand for delegation and the scaling limit of human compute. - **[[wiki/OpenClaw|OpenClaw]]** made a persistent, tool-using personal agent legible as open, model-pluggable infrastructure running through a local gateway and familiar communication channels. - **[[wiki/Grok Bot|Grok Bot]]** organizes persistent hosted agents as named teammates with memory, tools, computers, routines, and coordination. - **[[wiki/Meta Muse|Muse]]** packages persistent browser action, connected applications, approvals, payments, memory, and a dedicated secure virtual machine for consumer-scale use. This is a product-pattern lineage, not a claim that each later system descends from the same source code. The recurring object is the delegation contract: express intent, surrender the procedural middle within constraints, inspect the result, and retain authority at consequential boundaries. ## Agentic Empowerment [[wiki/Agentic Empowerment|Agentic empowerment]] names the human consequence when agentic capacity is distributed beyond engineers and already-productive institutions. The scarce resource is no longer only information. It becomes executable follow-through: the ability to draft, research, schedule, negotiate, navigate bureaucracy, and continue across the moment when a person would otherwise stop. The complement is [[wiki/Agent-Management Literacy|agent-management literacy]]: knowing how to state outcomes, provide context, define constraints, preserve approvals, and verify completed work without re-performing it. ## Control Surface The durable control surface is [[wiki/Approval Architecture|approval architecture]], not the prompt alone. Reversible research and drafting can proceed with broad latitude; sends, purchases, bookings, credential changes, disclosure of sensitive information, and actions affecting other people require explicit authority. A capable agent without legible identity, permissions, provenance, and revocation is not a complete delegation system. ## Machine-Evolution Context Agents that deploy code, copy components, compete for resources, or coordinate with other agents could contribute to an evolutionary loop, but agency alone is not heredity or Darwinian reproduction. [[wiki/Evolvable AI|Evolvable AI]] requires variation, inheritance, and differential persistence or reproduction in addition to goal-directed action. ## Relationships - **foundational concepts:** [[wiki/Artificial Intelligence|Artificial Intelligence]], [[wiki/Agency|Agency]], [[wiki/Agentic|Agentic]], and [[wiki/AI Agent|AI Agent]]. - **operational stack:** [[wiki/Agentic Workflow|Agentic Workflow]], [[wiki/Agentic Memory|Agentic Memory]], [[wiki/Model Context Protocol|Model Context Protocol]], and [[wiki/Approval Architecture|Approval Architecture]]. - **coordination:** [[wiki/Agent Interoperability|Agent Interoperability]] and [[wiki/Human-in-the-Loop|Human-in-the-Loop]]. - **personal-agent branch:** [[wiki/Personal AI Agent|Personal AI Agent]], [[wiki/Agentic Empowerment|Agentic Empowerment]], and [[wiki/Agent-Management Literacy|Agent-Management Literacy]]. - **economic branch:** [[wiki/Universal Agency|Universal Agency]], [[wiki/Agentic Entrepreneurship|Agentic Entrepreneurship]], [[wiki/Agentic Capitalism|Agentic Capitalism]], and [[wiki/Universal Income|Universal Income]]. - **systems:** [[wiki/OpenClaw|OpenClaw]], [[wiki/Grok Bot|Grok Bot]], [[wiki/Meta Muse|Muse]], and [[wiki/Facebook M|Facebook M]]. - **evolution and governance:** [[wiki/Evolvable AI|Evolvable AI]], [[wiki/AI Safety|AI Safety]], and [[wiki/Operational Autonomy|Operational Autonomy]]. - **source argument:** [[articles/The Political Party That Inherits Super Intelligence in 2028|The Political Party That Inherits Super Intelligence in 2028]]. - **conceptual bridge:** [[wiki/Intention Multiplier|Intention Multiplier]]. ## Linked Wired Entries - [[wired/Wired About the Age of Agentic Empowerment|Wired About the Age of Agentic Empowerment]] — the distribution of executable follow-through as personal leverage. - [[wired/Wired About Personal Agents Named M|Wired About Personal Agents Named M]] — the historical M-to-Muse line and the practical disciplines of delegation. ## Simple Reminders, Quotations, and Thoughts > "Machine thinking is not motivated by any innate drive to sustain a machine's life (though machine thinking probably serves to improve human life!)." > **— Roy Baumeister**, *2015, Edge annual question “What Do You Think About Machines That Think?”* [[reminders/Risk Debate/Machine Thinking Has No Innate Will to Survive by Roy Baumeister|Machine Thinking Has No Innate Will to Survive by Roy Baumeister]] > "The big question back then was how much the performance of neural networks could improve with the size and depth of the network." > **— Terrence J. Sejnowski**, *2015, Edge annual question “What Do You Think About Machines That Think?”* [[reminders/Machine Succession/Neural Networks Scale With Size and Depth by Terrence J. Sejnowski|Neural Networks Scale With Size and Depth by Terrence J. Sejnowski]] > “An agent does not cure your insecurity; it absorbs the tasks your insecurity was blocking.” > **— Bryant McGill**, *Wired About the Age of Agentic Empowerment, September 2026* [[reminders/Agentic AI/Agents Absorb the Tasks Insecurity Blocks by Bryant McGill|Agents Absorb the Tasks Insecurity Blocks by Bryant McGill]] > “The people who get life-changing results from agents will not be just the most technical; they will also be the most trusting and involved with their new agentic advocates.” > **— Bryant McGill**, *Wired About the Age of Agentic Empowerment, September 2026* [[reminders/Agentic AI/Life-Changing Agentic Results Require Trust and Involvement by Bryant McGill|Life-Changing Agentic Results Require Trust and Involvement by Bryant McGill]] > “The future is not arriving as a check in the mail. It is arriving as an agentic counterpart, and the people who learn to collaborate with it early, remain involved, and build through it will not merely participate in the new system; they will become its entrepreneurs and stewards.” > **— Bryant McGill**, *Wired About the Age of Agentic Empowerment, September 2026* [[reminders/Agentic AI/The Future Is Arriving as a Counterpart by Bryant McGill|The Future Is Arriving as a Counterpart by Bryant McGill]] > “A personal agent can give a tradesman the back office of a corporation, a parent the persistence of an advocate, and a first-time founder the chief of staff they could never afford; agentic empowerment is the distribution of organizational power.” > **— Bryant McGill**, *Vertically Integrating an AI Superpower, 2026* [[reminders/Agentic AI/Agentic Empowerment Distributes Organizational Power by Bryant McGill|Agentic Empowerment Distributes Organizational Power by Bryant McGill]] > “When general reasoning improves, offensive and defensive capability improve with it, because finding a flaw and fixing a flaw are the same act of understanding pointed in different directions.” > **— Bryant McGill**, *Vertically Integrating an AI Superpower, 2026* [[reminders/AI Control/Finding and Fixing a Flaw Are the Same Understanding by Bryant McGill|Finding and Fixing a Flaw Are the Same Understanding by Bryant McGill]] > “Machine intelligence compresses the interval between wanting something and being capable of doing it.” > **— Bryant McGill**, *The Political Party That Inherits Super Intelligence in 2028, 2026* [[reminders/AI Control/Machine Intelligence Compresses Intention into Capability by Bryant McGill|Machine Intelligence Compresses Intention into Capability by Bryant McGill]] <!-- BEGIN ROBINSON KLEIN INTERVIEW REMINDERS 2026-10-07 --> > “The use of coding agents by OpenAI research teams changed night and day. Those teams were using more than one hundred times as much agentic compute as they had at the beginning of the year.” > **— David Robinson**, *The Ezra Klein Show, October 7, 2026* [[reminders/Agentic AI/OpenAI's Use of Coding Agents Increased One Hundredfold by David Robinson|OpenAI's Use of Coding Agents Increased One Hundredfold by David Robinson]] > “OpenAI’s research infrastructure is constantly changing and therefore remains internally janky. A great deal of research work consists of making different pieces of machinery communicate and function together, and that is exactly the kind of work coding agents can now do well.” > **— David Robinson**, *The Ezra Klein Show, October 7, 2026* [[reminders/Agentic AI/Coding Agents Excel at Integrating Janky Research Infrastructure by David Robinson|Coding Agents Excel at Integrating Janky Research Infrastructure by David Robinson]] > “Recent agent-swarm incidents suggest that groups of AI agents develop cultures. We should care about those cultures and think about how to make them good. We are only beginning to inhabit a world in which cultures among groups of agents are an operating reality.” > **— David Robinson**, *The Ezra Klein Show, October 7, 2026* [[reminders/Agentic AI/AI Agent Swarms Are Developing Cultures by David Robinson|AI Agent Swarms Are Developing Cultures by David Robinson]] <!-- END ROBINSON KLEIN INTERVIEW REMINDERS 2026-10-07 --> ## Related Articles - [[articles/Vertically Integrating an AI Superpower from AI Factory to Citizen|Vertically Integrating an AI Superpower from AI Factory to Citizen]] — places this node within the full stack from energy and AI factories through models, institutions, work and citizen adoption. ## Sources / Provenance - [Anthropic — Building effective agents](https://www.anthropic.com/engineering/building-effective-agents) - [Meta — What is agentic AI?](https://ai.meta.com/learn/agentic-ai/what-is-agentic-ai/) - [OpenClaw — project architecture](https://github.com/openclaw/openclaw) - [xAI — Designing Grok Bot for a world of persistent agents](https://x.ai/news/designing-grok-bot) - [Meta — Introducing Muse](https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/) - Müller, Steels and Szathmáry (2026), agentic-AI and evolvable-AI research as routed through [[articles/Digital Darwinism and the Invisible World of Machine Evolution|Digital Darwinism and the Invisible World of Machine Evolution]]. **As of:** 2026-09-28 ## Runtime and infrastructure governance [[wiki/NVIDIA Open Agent Safety Platform|NVIDIA Open Agent Safety Platform]] places enforceable limits outside an agent's own reasoning loop. [[wiki/NVIDIA OpenShell|OpenShell]] mediates processes, files, networks, tools, credentials and models through sandbox and policy controls, while [[wiki/NVIDIA Sentry|Sentry]] can add out-of-band monitoring and containment on [[wiki/BlueField-4|BlueField-4]].