The Approval Workflow: Why We Never Auto-Execute

When we tell people that AI agents can manage their ad campaigns, the first question is always: "But what if it does something wrong?"

It's a fair question. An agent with access to your ad accounts could theoretically burn through your budget on a poorly targeted campaign. That's why every action in Headless HQ goes through an approval workflow.

Three approval modes

Full approval is the default. Every action the agent proposes — creating a campaign, adjusting a budget, publishing a creative — shows up in your approval queue with a clear explanation of what will happen and why. You review and approve or reject with one click.

Threshold mode lets you set a dollar amount. Actions below the threshold are auto-approved; actions above it require your review. This is useful once you trust the agent's judgment on routine tasks but still want oversight on larger decisions.

Autonomous mode auto-approves everything. We include it for completeness, but we recommend working up to it gradually.

The anatomy of an approval

Each approval request includes:

• The specific action to be taken (e.g., "Create Meta campaign targeting 25-34 year olds in US/CA")
• The agent's reasoning (e.g., "Based on Q4 performance data, this demographic had the highest ROAS")
• Estimated impact (e.g., "Projected daily spend: $150, estimated reach: 45K")
• A diff view showing exactly what will change

You get real-time notifications via WebSocket, email, or both. Pending approvals never expire — they wait until you act on them.

Why not just auto-execute?

We could build a system that just does things. But trust is earned, not assumed. The approval layer lets you build confidence in the agent's judgment over time, at your own pace. It also creates an audit trail — every action taken on your behalf is logged with the reasoning behind it.