A semi-autonomous AI marketing team, running the growth of one practice
A five-agent growth agency, running my own practice
Five AI agents — a Growth CEO, a trend researcher, a content writer, a distributor, and a performance analyst — running BetterStory's marketing on their own, while I approve strategy instead of writing every post.
An agency retainer or a one-person marketing scramble → a small, named team that runs the work itself.
| Before | After | |
|---|---|---|
| Marketing for a one-person practice | an agency retainer, or done in the gaps between client work | a named five-agent team running it as the work |
| Cadence | sporadic, when you have a Friday afternoon | weekly content, daily distribution, monthly review |
| Trend awareness | 'I'll catch up on the field later' | a Scout agent reading the field on its own |
| What gets posted | whatever you remembered to write | what the CEO and Scout agreed was worth saying |
| Your role | do every step yourself, or hand it all over | approve strategy, set budgets, review output |
| Cost | an agency retainer or a part-time hire | model calls and the orchestration runtime |
What normally takes an agency retainer or a part-time marketer (whichever you can afford to lose) becomes a named team running 24/7, in your voice, on your strategy, for the cost of the model calls. Marketing stops being the thing that doesn't get done — it becomes a function of the practice, with someone responsible for each piece.
What I built
A semi-autonomous AI marketing team for BetterStory — my own practice — running on the multi-agent orchestration platform I work on (Paperclip). Five agents in a clear org chart, each with a real role:
- Growth CEO. Owns the content calendar and the quarterly strategy. Reviews work from the other agents, reports to me, and adjusts based on what Intel sees.
- Scout. Reads what's actually getting traction in AI-native consulting and adjacent fields — papers, posts, threads — and feeds the CEO a short weekly digest of what's worth talking about.
- Forge. Drafts content in my voice against the calendar — essays, threads, replies — with a tight style spec that prevents drift.
- Distro. Publishes the work across X and Reddit on the right cadence and in the right framing per channel.
- Intel. Watches SEO and performance, reports what landed and what didn't, and feeds the CEO so next week's calendar reflects what actually works.
Every agent has a bounded role, a budget the CEO enforces, and a reporting line that lands on my desk weekly. I approve strategy, override when I disagree, and audit output. I don't write every post.
Why it matters
This is the showcase running on its own scaffolding. The pitch of "an autonomous AI company" is easy to make in a demo and hard to make in production — most multi-agent systems work right up until they drift off-voice, blow a budget, or quietly stop producing. The constraint here is the org chart: a CEO that owns voice and strategy, agents with named roles and bounded budgets, a board (me) that approves work above a threshold. It's how a real company stops a single rogue employee from breaking the brand — and it works for the same reason in software.
The reason this case exists at all is the breadth proof: I'm not just recommending AI-native systems to my clients — I'm running my own practice on one, watching what breaks, and shipping the fixes back into the platform. The marketing of BetterStory is the test bed.
- 01CEO sets the strategy
The Growth CEO agent maintains a content calendar and a quarterly strategy, both based on the goals I set and the performance Intel reports back.
- 02Scout finds the trends
A Scout agent reads what's actually getting traction in AI-native consulting and adjacent fields — papers, posts, threads — and feeds the CEO a short weekly digest.
- 03Forge writes the content
A Forge agent drafts the content in my voice — essays, threads, replies — against the calendar the CEO set.
- 04Distro publishes the work
A Distro agent ships the content across X and Reddit on the right cadence, with the right framing per channel.
- 05Intel measures the result
An Intel agent watches SEO and performance, reports what landed and what didn't, and feeds the CEO so next week's calendar reflects what actually works.
- 06I approve strategy
I review strategy, override when I disagree, set budgets per agent, and audit output — I don't write every post.
Marketing for a one-person practice becomes the work of a named five-agent team instead of a recurring guilty thing. The CEO sets strategy, Scout reads the field, Forge writes in your voice, Distro publishes, Intel measures — and I approve strategy and set budgets, instead of writing every post.