Holdcos talk governance. Mid-market still prompts.

Two agencies can say “we use AI” in the same pitch and mean opposite machines.

At the top of the holding groups, the language has already moved: brand controls, approved models, review workflows, responsible-use PDFs. Some of that is real. Some of it is a committee covering a generator. Either way, the sentence exists.

In the mid-market — the independents who actually run most local and specialist brands — the sentence is still a ChatGPT window and a junior with a deadline. That is not a moral failing. It is a stack failing. Brand governance AI was sold as an enterprise SKU. Independents were told to prompt harder.

This is the gap. Not talent. Not taste. Whether governance is a document or a loop.

PDFs do not refuse a post

A policy that cannot stop a publish is a press release. NIST’s AI RMF is useful because it is operational: map context, measure, manage. A twelve-page “AI principles” deck that never touches the scheduler is costume.

The mid-market problem is the inverse. There is no deck. There is a model, a paste, a calendar. Fluency is not authority. The copy passes. The brand thins. The principal finds out when the client does.

OECD still wants a human accountable. ACCC still wants claims you can prove. Neither regulator asks whether you are WPP or a twelve-person shop in Brisbane. The feed does not care either.

What the independents actually need

Not a holdco committee. Not another generator. The scarce goods we named when content got cheap: judgement, integrity, control, proof — at a price and an operating load a mid-market roster can carry.

That is an online agentic closed loop, not a values page:

• Lions Law as intake, not a poster
• Permission by goal, written
• So What Law as a refuse
• A name on publish
• Evidence grades, not a vanity total

Qy Partners is the proof test for that, and the programme page is the working form. If a holdco already has this as software, good — use it. If an independent only has prompts, do not pretend the PDF will arrive later. Build the record now, on the brands you already have.

The embarrassing truth: a twelve-person shop with a closed loop is more grown-up than a network with a governance theatre and a prompt window on the account team. Clients will learn to tell the difference. Some already have.

What to ask in a pitch, either side of the table

If you are the client: “Show me a post you refused this month, and the rule that killed it.” If they cannot, you are buying fluency.

If you are the independent: do not compete with holdco adjectives. Compete on artefacts. Brand record, permission slip, named human, graded month. That set is rarer than a generator licence, and it is the only set Qy Partners cares about.

The audit at the end of this series is how you score yourself before they do.

Restart, if your own mid-career is the mid-market: gailmacleod.ai. Soft door: app.theagencyiq.ai.

Frequently asked questions

Do holding companies actually have better AI governance?

Sometimes the sentence exists. Sometimes it is a PDF. Ask whether the policy can refuse a publish on a live account. If it cannot, it is theatre.

What should a mid-market agency do instead of prompting?

Install a closed loop: brand truth in, permission by goal, grounded claims, named accountability, graded evidence. Tools are downstream of that.

Is this only for big retainers?

No. A small roster is where leakage hurts most. Boutique voice becoming franchise voice is a mid-market failure, not a network one.