{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreihjzo6k4u5god2sriysralapgtrqkh7pafpcdubcfib3slyihsueu",
    "uri": "at://did:plc:25rdn5elo5izoxrmtis34zuk/app.bsky.feed.post/3mpszame7tbj2"
  },
  "path": "/claireg/part-2-agentic-ai-4k9p",
  "publishedAt": "2026-07-04T11:53:30.000Z",
  "site": "https://dev.to",
  "tags": [
    "agents",
    "ai",
    "automation",
    "llm"
  ],
  "textContent": "This is where the real confusion — and the real governance problem — actually lives. People talk about “AI deciding,” “AI acting,” “AI refusing,” “AI escalating,” “AI breaking rules,” “AI needing governance”… None of that belongs to Functional AI. It belongs here.\n\nAgentic AI is the category everyone argues about, but almost nobody defines correctly.\n\n##  What Agentic AI is\n\nAgentic AI is action producing machinery.\n\nIt is any system that can:\n\n  * take actions\n  * call tools\n  * change states\n  * execute workflows\n  * trigger processes\n  * affect the world\n\n\n\nAgentic AI includes:\n\n  * tool calling LLMs\n  * workflow agents\n  * autonomous loops\n  * multi agent systems\n  * planning + execution systems\n  * “AI assistants” that actually do things\n\n\n\nAgentic AI is technically not a model. It is a system built around a model.\n\nIt is the first category of AI that can:\n\n  * initiate\n  * escalate\n  * choose between options\n  * affect external systems\n  * cause consequences\n\n\n\nThis is why governance belongs here — not in Functional AI.\n\n##  What Agentic AI is not\n\nAgentic AI is not magic. It is not a mind. It is not a sovereign. It is not a decision maker in the human sense.\n\nAgentic AI is not:\n\n  * self aware\n  * self directed\n  * goal seeking in a human way\n  * capable of interpreting legitimacy\n  * capable of understanding constraints\n  * capable of moral reasoning\n  * capable of “knowing” what it is doing\n\n\n\nPeople project agency onto systems that simulate agency.\n\nAgentic AI does not “want.” It does not “intend.” It does not “understand rules.” It does not “respect authority.”\n\nIt executes patterns inside a wrapper that looks like agency.\n\n##  The Domain Layer (scope, not architecture)\n\nAgentic AI becomes “Domain Agents” when deployed inside specific fields:\n\n  * medical triage agents\n  * legal drafting agents\n  * financial decision agents\n  * aviation workflow agents\n  * industrial automation agents\n\n\n\nBut this does not change the system type.\n\nIt is still Agentic AI — just operating inside a domain.\n\nDomain context affects:\n\n  * risk\n  * consequences\n  * escalation paths\n  * authority layers\n  * governance requirements\n\n\n\nBut it does not give the agent real understanding or real intent.\n\n##  How the Human Authority Layers attach\n\nThis is where things get serious.\n\nAgentic AI is the **first** system type that touches all three human authority layers.\n\n**Regulated AI (legal ecosystem)**\n\nStrong attachment. Regulators care about:\n\n  * actions\n  * consequences\n  * auditability\n  * accountability\n  * risk classification\n  * compliance in execution\n\n\n\nBecause Agentic AI can do _things_ , legal exposure is real.\n\n**Responsible AI (ethical ecosystem)**\n\nModerate attachment. Ethics people worry about:\n\n  * fairness in decisions\n  * bias in actions\n  * transparency in workflows\n  * explainability of choices\n  * But ethics alone cannot govern actions.\n  * Human Legitimacy (political ecosystem)\n  * Maximum attachment. This is the layer everyone forgets.\n  * Agentic AI raises questions like:\n  * Who authorises the agent?\n  * Who approves its actions?\n  * Who sets its constraints?\n  * Who does it escalate to?\n  * Who is accountable for its behaviour?\n\n\n\nThis is **governance** , **not ethics.**\n\nThis is **authority** , **not transparency**.\n\nThis is **legitimacy** , **not fairness**.\n\nGovernance belongs here — not in **Functional AI**.\n\n##  The Tribes Who Worry About Agentic AI\n\nThis is where the sociology shifts.\n\nThe people who worry about Agentic AI are:\n\n  * governance experts\n  * risk officers\n  * compliance teams\n  * regulators\n  * safety engineers\n  * operations leaders\n  * enterprise architects\n  * political theorists\n  * security professionals\n\n\n\n**Their fear:** “The agent will take an action it shouldn’t.”\n\n**Their behaviour:** They try to impose governance on systems that do not understand governance.\n\n**Their noise contribution:** They treat agent wrappers as if they were minds.\n\n##  Vendor Incentives\n\nThis is where vendor confusion becomes **dangerous**.\n\nVendors pitch:\n\n  * agent orchestration\n  * workflow automation\n  * agent governance\n  * approval systems\n  * safety rails\n  * “enterprise agent platforms”\n\n\n\nBut they often describe these as:\n\n  * “AI governance”\n  * “AI safety”\n  * “AI trust”\n  * “AI compliance”\n  * “AI productivity”\n\n\n\nUsers don’t know the difference between:\n\n  * a model\n  * an agent\n  * an action\n  * a workflow\n  * a constraint\n  * a decision boundary\n\n\n\nSo vendors collapse everything together.\n\n**The problem** : Most of these tools assume agents can interpret constraints and understand authorization — which they cannot.\n\n##  The Noise Layer\n\nAgentic AI is where the public panic lives.\n\nThe noise includes:\n\n  * “AI decided”\n  * “AI refused”\n  * “AI broke the rule”\n  * “AI escalated incorrectly”\n  * “AI acted dangerously”\n  * “AI needs governance”\n  * “AI needs authority”\n\n\n\nAll of this is **category collapse**.\n\nPeople treat agent wrappers as if they were autonomous minds.\n\nAgentic AI is not a **sovereign**.\nIt is not a **decision maker**.\nIt is not a **moral actor**.\nIt is not a **political entity**.\n\nIt is a system that **executes patterns inside an action loop**.\n\n##  The Clean Takeaway\n\nAgentic AI = **action system**.\n\nIf you treat it like a mind, you will:\n\n  * govern it wrong\n  * regulate it wrong\n  * design it wrong\n  * panic about the wrong things\n  * ignore the real risks\n  * collapse categories **- hurt Claire’s senses**\n\n\n\nAgentic AI is the first system type that can act — and the most misclassified.",
  "title": "Part 2 - Agentic AI"
}