# Agent governance

## Agentic marketing, governed by your guardrails

Set what each agent can do, where it can act, and when it must ask. Nothing runs outside those rules.

**Sensitive action controls** Allow, ask, or block budget, targeting, and campaign changes.

**Autonomy with clear boundaries** Scope accounts, budget movement, and manager.

**A complete decision trail** See the agent, manager, approver, and platform result.

## Why governance matters

### A confident answer can still be the wrong action.

When AI can touch live spend, a vague instruction is not a control. High-stakes decisions need an explicit policy before an agent can move money or change a campaign.

**A small error can compound quickly** A budget or targeting mistake can spread across live campaigns before the team sees the consequence.

**Intent is not the same as permission** Knowing what the team wants does not define which actions an agent may take without approval.

**Human error compounds at agent speed** A mistaken Allow rule can repeat one human error across every action until someone changes it.

Agents stay adaptive. Your controls travel with every action.

## Every action gets a safety switch.

Give every agent the right level of autonomy for budgets, targeting, campaign state, and every other live action.

**Budget Pacer** Managed by Evan Brooks

### Action permissions

**Change budgets** Inside the budget cap  
AllowAskBlock

**Pause campaigns** Only in scoped accounts  
AllowAskBlock

**Edit targeting** Never runs  
AllowAskBlock

**Turn ads on or off** Inside assigned scope  
AllowAskBlock

**$3B+** annualized spend under agent management  
**14,000+** Meta and Google ad accounts connected

**Budget Pacer** Active  
Scoped accounts **Superco USBrand Search**  
Manager **Evan Brooks**  
Max budget change **20% per action**

## Keep every action inside hard limits.

Choose the accounts, action list, budget ceiling, and manager before an agent can touch live spend.

**Scope controls access** An agent can only reach the accounts and actions assigned to its scope.

**Every agent has an assigned manager** If the manager leaves, the agent pauses until a new one is assigned.

**Work is reported where your team works** Agents notify the team in Slack or email when work is completed, needs approval, or cannot run.

## Fleet management

### Manage the fleet by account.

Filter every agent by account scope, then see its owner and latest run in one place.

**Agent fleet** 4 shownPause all

All accountsSuperco USMeta AcquisitionGoogle Search

AgentAccount scopeOwnerLast run

**Budget Pacer** Daily  
Superco US  
**Evan Brooks**  
20m ago

**Creative Fatigue Watcher** Daily  
Meta Acquisition  
**Maya Chen**  
8m ago

**Account Auditor** Weekly  
Google Search  
**Nia Carter**  
1h ago

**Weekly Client Report** Weekly  
All accounts  
**Evan Brooks**  
8d ago

## Every action leaves a trail.

Task Logs show the agent, its manager, the approver, the platform result, and the reason behind every live change.

**Task Logs** Search by task, agent, account, or platform

FilterAgent: Budget PacerStatus: All

TaskCreatedPlatformStatusAgent

**Reduce budget on Prospecting Q3** Today, 8:01 AMMetaExecutedBudget Pacer›

**Decision trail:** Approved by **Evan Brooks**Managed by **Maya Chen**Applied through **GoMarble**

**Pause Broad US ad set** Today, 9:40 AMGoogle AdsNeeds reviewCreative Fatigue Watcher›

**Increase bid on running shoes** 6 Jul, 2:15 PMGoogle AdsFailedCPA Spike Alert›

## From answer to accountable action

### AI Chatbots need guardrails

Claude and ChatGPT execute blindly without guardrails, but prompts are not controls. GoMarble adds hard action policies and a traceable execution layer between a recommendation and a live account change.

### Governance starts with platform access.

GoMarble connects through official platform consent, respects the access each platform grants, and applies your agent rules inside that ceiling.