Permission slips by goal: what agents may do — and may not

“Grow the brand” is not a brief. It is a shrug.

An agent given a shrug will fill the calendar anyway. That is not initiative. That is the absence of a permission slip. In an online agentic closed loop, step two is brutal on purpose: goal → permission. The agent may only act inside a named goal with named bounds. Everything else is a refuse.

This is the piece agencies skip because it feels like legal, and legal feels like slow. It is slower than a prompt. It is faster than a brand crisis, a client resignation, or an ACCC letter about a claim nobody can prove.

A goal is not permission

A goal says what you want. Permission says what you are allowed to do in order to want it.

Book consults: named offer, hours, locations — not invented urgency or clinical claims you cannot prove. Hire two seniors: role and salary you will actually pay — not “we’re a family” or diversity claims without practice. Launch a menu: dishes that exist, prices that are current — not “best in the city” without grounds. Rebuild trust: what you changed — not a joke reel that pretends it did not happen.

If the forbidden column can still ship, you do not have agentic AI for digital agencies. You have a caption factory with a slogan.

Lions Law tells the system who the brand is. The permission slip tells it what this week’s job is allowed to cost the brand. A brand can be true and still do a stupid thing in pursuit of a KPI.

How to write one in an hour

One slip per live goal. A page, not a novel.

1. Goal in one sentence a client would sign.
2. Success evidence — what would count, with a grade, not a like count.
3. Allowed acts — channels, formats, verbs, claims, offers.
4. Forbidden acts — vulgarly specific.
5. Escalations — what must come back to a human even in Auto.
6. Expiry — permission dies when the goal dies.

The OECD AI Principles put accountability on human actors. The slip is how you make that real before 7 a.m. Auto. NIST wants context mapped; this is the map for a campaign, not the constitution.

So What Law still sits on every act: permitted and empty is still a refuse.

Manual and Auto under the same slip

Auto is extra speed inside the same bounds — not extra permission. Brand rules lock before publish whether a human taps or the calendar does. If you would not give a junior Auto on this goal, do not give it to the agent.

Human accountability is the name on the slip. Qy Partners only works if each client brand has its own slips. Shared permission across a roster is how boutique voice becomes franchise voice.

If you cannot write the forbidden column, you are not ready to generate. Write it. Then generate inside it.

A worked refuse, so this is not abstract: a dental group wants “more implants this quarter.” Permission may include the consult offer, hours, locations, and “book.” It may not include before-and-afters you do not have consent for, pain-free guarantees, or “from $X” if $X is not a price you will honour. The agent that writes the guarantee anyway is not bold. It is ungoverned. Kill it. File the kill. That file is how you teach the next goal.

Kitchen-table cousin: a Restart brief is a permission slip for your next five years. gailmacleod.ai. Soft door: app.theagencyiq.ai.

Frequently asked questions

What is a permission slip by goal?

A one-page bound on a live goal: allowed acts, forbidden acts, evidence that would count, who escalates, and when permission expires.

Can’t brand rules cover this?

Brand rules are the constitution (Lions Law). A permission slip is this week’s job. You need both.

Does Auto-publish need a different slip?

No. Same slip. Auto changes who taps, not what is allowed.