Creative test loop agent | GTM AI agents

What it does

Creative fatigue is the quiet tax on every paid account: the winning ad decays, nobody has a replacement ready, and the account coasts on a fading asset. Testing consistently requires production discipline that most teams cannot sustain by hand, week after week.

This agent runs the loop. It generates variants inside your visual system, proposes each enrollment as an Approve, Modify or Skip card in Slack, runs approved creatives in a dedicated test set on a hard budget cap, and judges them with a deterministic gate: cost per lead against your graduation threshold. Winners are recreated in the scale set; losers are paused before they get expensive.

The loop also protects the scale set itself. Active-ad capacity is capped, and when a graduate arrives the worst incumbent rotates out, highest cost first. Every verdict is kept as memory, so next week's variants start from what this week proved.

Before and after

Before

New creative ships when someone finds a free afternoon
Tests run on whatever budget was left over
Winners are declared by eyeballing the dashboard
The scale campaign accumulates tired ads

With the agent

A steady pipeline of on-brand variants, every week
Each test capped at a fixed budget, never more
Graduation is a deterministic cost gate, not a vibe
The scale set stays at capacity with its best performers

How it works

Seven gates. The approval card is the front door, the budget cap is the ceiling, and the deterministic verdict is the exit.

1
Generate variants on brand
New creatives come out of your visual system and the accumulated learnings of past rounds, not from a random prompt.
2
Propose before spending
Each candidate arrives in Slack as an Approve, Modify or Skip card. Pierce proposes, you approve: nothing enters the account without a human decision.
3
Test on a hard budget cap
Approved creatives run in a dedicated test ad set, each with a fixed test budget it cannot exceed. A bad creative has a known, small, maximum cost.
4
Judge with a deterministic gate
Cost per lead over the test window against your graduation threshold. Beat it and graduate; miss it and pause. No committee, no eyeballing.
5
Graduate winners into the scale set
Winners are recreated in the scale ad set. The test copy is paused, not deleted, so the full history stays auditable.
6
Evict the worst to keep capacity
The scale set has a maximum of active ads. When a graduate arrives over capacity, the highest-cost incumbents rotate out first.
7
Keep the learnings
Every verdict persists as memory: which angles, formats and hooks graduated. The next generation round starts from evidence.