What an online agentic closed loop actually is

Most of what agencies bought as “AI” in the last two years is a caption factory with a login.

A model that can write five Instagram lines before the kettle boils is not an agent. A scheduler that fires those lines into a feed is not a loop. A dashboard that counts likes is not learning. String them together and you still do not have an online agentic closed loop. You have volume with a hangover.

Gail Macleod defines it this way, on the record, for communications and digital agencies: brand-trained agents that plan, generate, publish, attribute, and learn without transferring brand authority to a fluent model.

That sentence is the whole argument. Everything else is how you refuse to fake it.

Fluency is not authority

The industry’s current disappointment is not mysterious. Teams prompted a fluent model, got fluent copy, published it, and watched the brand get a little less itself. Clients felt it before the agency did. The work sounded like everyone else’s work because it was trained on everyone else’s work.

Fluency is the ability to produce language that passes. Authority is the right to say what the brand means — and the duty to stop what it must not say. Those are different goods. Confusing them is how you get a feed that is never quite wrong and never quite yours.

The NIST AI Risk Management Framework exists, in part, because organisations kept treating output quality as if it were governance. It is not. The OECD AI Principles (2019, updated 2024) still put accountability on human actors. A model does not own a brand. A principal does.

If you need the short version of the disappointment, it is Fluency Is Not Authority. This piece is the system that disappointment was pointing at.

Seven moves, then back again

An online agentic closed loop is not a feature list. It is a sequence that is not allowed to skip.

1. Brand truth in (Lions Law). You may not agree with AI on your brand — have you trained it to understand your brand? Humans and machines must recognise the same brand truth. If the agent cannot state the promise, the bans, the audience, and the claims that are allowed, it is not trained. It is prompted. Lions Law is the intake, not a poster on the wall.

2. Goal → permission. A goal without a permission slip is a wish. “Grow awareness” is not permission to invent a testimonial. “Book consults” is not permission to discount a regulated service. The agent may only act inside a named goal with named bounds. That is permission slips by goal.

3. Generate inside rules (grounded claims only). Drafts are cheap. Unsubstantiated claims are expensive. In Australia, the ACCC is blunt: claims should be true, accurate, and based on reasonable grounds; a business must be able to prove what it advertises. Generation that cannot point at a source does not wait for legal. It dies in the gate.

4. Publish with human accountability (AI does not own the brand). Manual or Auto is a lever, not a moral holiday. Brand rules lock before publish either way. The human still owns the brand when the calendar fires at 7 a.m. See human accountability when agents publish and brand rules lock before publish.

5. Measure honestly (evidence grades; likes ≠ proof). Likes are a weather report. They are not a verdict on strategy. A loop that treats engagement as truth will teach the brand to become a joke account. Evidence has grades. Why likes corrupt learning.

6. Learn without hijacking brand truth. The system may propose. It may not rewrite the constitution because a reel did well. Observation is not authority.

7. Back to brand truth. The next plan starts from the same truth the last one was not allowed to abandon. That is the closed in closed loop. Open loops accumulate drift. Closed loops accumulate judgement.

So What Law gates every shippable act. If the buyer cannot answer so what, it does not ship. Empty publish is not a content strategy. It is a leak. So What Law.

What it is not

It is not a plugin that pastes on-brand adjectives into a box. It is not Auto-post as a personality. It is not a holdco PDF titled governance while the mid-market still lives in a prompt window. That split is Holdcos talk governance. Mid-market still prompts.. It is not transferring the brand to the most fluent model in the room.

A caption factory optimises for output. A closed loop optimises for permitted output that can be attributed, graded, and used as evidence without rewriting who the brand is.

What a principal must be able to show

If a client asks how you use AI on their brand, a paragraph about “our process” is not an answer. You need a record:

• the brand truth the work was generated from
• the goal and the permission that sat on that goal
• what was refused, not only what was posted
• who was accountable at publish — Manual or Auto, same rules
• evidence with a grade, not a vanity total
• what the system is allowed to learn, and what it is forbidden to become

That is the Qy Partners test. What agency principals must prove. The measurement record is the public shape of honest numbers. The audit is how you score your own shop before a client does.

TheAgencyIQ’s Qy! runs this pattern for social across Facebook, Instagram, LinkedIn, X and YouTube: brand rules lock before publish; Manual or Auto; the human owns the brand. Agencies see it as Qy Partners. None of that is the definition. The definition is the loop. The product is one implementation. How to use it is operations. This page is doctrine.

The founder path — Stratcom discipline into agentic systems, packaging DNA as biography not as a customer list — sits in From Stratcom discipline to agentic systems and the founder hub. Restart for over-50s and the kitchen-table version of So What Law lives at gailmacleod.ai.

If you only remember one thing: do not confuse a fluent model with a governed loop. Train it. Bound it. Publish with a name on the work. Measure without lying. Learn without hijacking. Return to brand truth. Then do it again.

Soft door, not the point of the piece: app.theagencyiq.ai.

Frequently asked questions

What is an online agentic closed loop?

Brand-trained agents that plan, generate, publish, attribute, and learn without transferring brand authority to a fluent model. Seven moves, then back to brand truth. So What Law gates every shippable act.

How is that different from ChatGPT plus a scheduler?

A prompt is not training. A schedule is not permission. Likes are not evidence. Without brand truth in, a goal-bound permission slip, grounded claims, human accountability at publish, and learning that cannot hijack the brand, you have a caption factory.

Does Auto-publish mean the AI owns the brand?

No. Manual or Auto is a lever. Brand rules lock before publish either way. AI does not own the brand. The human does.