AI Growth Marketer
Goal, Brief and Memory
The agent starts with marketing expertise: how these platforms behave, what usually works, where the usual mistakes are. Memory is what it adds to that from your account — every test run, every result, every call you made — so over time it is not applying general practice to your business, it is applying what has been shown to work for it.
That knowledge base is what makes the work compound. A new request starts from what was learned last month rather than from scratch, and the proposals get better the longer it runs.
Think of it the way you would think about hiring someone good. They arrive knowing the craft, but they are not useful on day one until they know your margins, your seasonality, the lines you will not cross and what you have already tried. The more of that you tell them, the faster they are worth the money. Same here — and unlike a hire, none of it walks out of the door.
The goal
Everything starts from what you are trying to buy and what you are willing to pay for it: orders, subscriptions, signups, waitlist places. That target is what the agent optimises toward and what it reports against, and it belongs in the brief so every piece of work is judged the same way.
The brief
The brief is how you give that context deliberately, rather than waiting for the agent to infer it.
Your workspace has one brief covering the business as a whole: what you sell, what you are trying to do, what you spend, how performance should be judged, and the rules that are not up for debate. Every request, workflow and co-pilot starts from it.
The context every request, workflow and co-pilot starts from.
- Business
- Direct-to-consumer jewellery, mostly US. Average order $140, repeat rate 22%. Gross margin 61%.
- Goals
- Grow orders 30% this quarter without pushing cost per order above $40. Emerald and Paraíba lines first.
- Budget
- $120,000 a month across paid. Meta and Google carry most of it; hold Reddit at a test.
- Measurement
- Judge on Shopify orders, not platform conversions. Incrementality test results override the platform view.
- Notes
- No discounting above 15%. Never use competitor names in copy. Creators must show the product being worn.
Each co-pilot has its own brief on top of that, narrowing it to a scope: the channel or area it owns, its target, and the limits it works inside. See co-pilots.
Briefs are worth revisiting. When the goal changes, or a constraint stops applying, changing the brief changes what the agent does next — no retraining and no re-explaining.
What it remembers
- What happened. Changes it made or proposed, when, and what followed.
- What worked. Which creatives, hooks, audiences and keywords performed, and where.
- How you decide. What you approved, what you rejected, and any reasons you gave.
- How your business works. Products, margins, seasonality and constraints you have told it.
What it does with that
A co-pilot uses memory to stop re-proposing what you already rejected. Creative work uses it to brief the next round from what won. Reporting uses it to explain a change rather than restate a number.
Where it lives
Memory belongs to your workspace. Each brand, account or client is separate, and nothing crosses between them.
On enterprise terms, your data is not used to train any cross-account model or AI agent, including in de-identified form. See trust and security.