Execution Agent
Not a recommendation. A change. Shipped on-platform with guardrails and an audit trail.
Every other agent's output converges here. Planning, Creative, and Analyst queue approved changes overnight — Execution pushes them to Meta, Google, TikTok, and LinkedIn in one coordinated batch. Guardrails enforce the limits. The audit trail records every move.
Signals In
Proposed changes from every other agent. Budget reallocations, creative pauses, test launches, competitive responses — each one carries the action, the signal behind it, and the proposing agent's confidence.
The configured guardrail policy. Daily volume limits, spend caps, blast-radius rules, structural-change cooldowns, auto-approve thresholds — all configurable per workspace.
The approval queue state. What's pending, who can approve, what the SLA is.
Live platform API state. Account health, rate limits, structural constraints — so changes that would fail at the platform get caught before they ship.
Actions Out
Changes shipped on-platform. Cut a budget. Launch a campaign. Swap a fatigued creative. Pause a campaign with a stale audience. Push a Lookalike increase off MMM headroom. All written directly to Meta, Google, TikTok, LinkedIn — no copy-pasting from a recommendation card to ad manager.
A live change log. Every change carries the proposing agent, the triggering signal, the approver, the API response, and the reversal path. Queryable in plain English, tamper-evident.
An approval queue. Pending changes with full context attached. Shipped changes with one-click rollback.
Reversal at any time. Every change has a one-click rollback — and the rollback is itself logged.
Questions
