Built an internal AI growth-operations console (Next.js 16 App Router, TypeScript, Tailwind) that runs the full weekly loop: client intake → AI research → strategy → creative pack → landing-page variants → human approval → export → performance → learnings → next backlog
AI Growth OperatorMarketing Ops
Our own AI growth engine: from client data to reviewed, export-ready experiments — every week, with a human in the loop.
Marketing & Growth Services (Shopify / D2C ecommerce)
aigrowthoperator.app
Workflow
- Intake
- Extract
- Generate
- Review
- Publish
project.pdf
fields.json
listing ✓
Human approval gate before publish
The challenge
Running weekly growth experiments for Shopify and D2C brands was bottlenecked by manual agency labor — research, strategy, creative, and landing-page variants were handcrafted per client, so output stayed low and learnings rarely fed the next batch. We needed a system to ship more high-quality, on-brand experiments each week without scaling headcount linearly.
Marketing Ops
Marketing & Growth Services (Shopify / D2C ecommerce)
What we built
The system, in parts.
Designed a structured Postgres data model in Drizzle (clients, brand profiles, products, daily metrics, competitor pages, experiments, creative & landing-page variants, approvals, publish jobs, performance snapshots, learnings)
Architected five typed AI agents (research, strategy, creative, landing-page, learning) where every run is logged to an agent_runs ledger with input/output JSON, model used, status, and errors for full auditability
Enforced a hard approval gate: nothing can be exported or published without an approved approval record, with audit logs on every decision and automatic flagging of high-risk health/beauty/finance/legal/medical claims
Shipped export-only V0 publishing (Markdown creative packs, HTML landing pages) so operators stay in control, with a Shopify-draft connector scoped for V1
Made the per-client workflow state visible in an operator dashboard (active clients, open runs, pending approvals, workflow timeline) so the team sees every experiment’s status at a glance
Outcomes
What changed for them.
Turns days of manual growth-agency work into a structured weekly loop aimed at a complete experiment pack per client in hours, not days
More high-quality, on-brand experiments shipped per week without proportional headcount growth
Every AI output is reviewable and gated — no creative or landing page reaches a client channel without human approval
Risky claims (health, beauty, finance, legal, medical) are flagged before they ever leave the building
Full audit trail: every AI run, approval, and export is logged, so the team can debug, defend, and improve
Learnings from past performance feed the next experiment backlog instead of being lost between sprints
How it’s built
The stack.
Next.js 16 (App Router)React 19TypeScript (strict)Tailwind CSS v4Drizzle ORMSupabase / PostgresZodVercel AI SDK
More work
See allTwo ways to start
Want a system like this?
Tell us the workflow you want to run itself. We will scope a focused first project — designed, built, and operated, with humans in control.