Managed agent team
A Scrum Master coordinates six channel and business specialists through shared priorities, context, and operating rules.
AI OPERATING LAB · KiwiFamilies.co.nz
Building an AI operating system inside a real business.
Liz uses KiwiFamilies.co.nz to test AI ideas under real operating conditions, develop safe practices, and prove measurable business value.
- Role
- Owner and Operator
- Model
- Sole operator with seven specialized AI agents
- Technology
- Generative AI APIs
- Timeline
- 2020 to present
30K+monthly readers
69Kpeak monthly pageviews with 2.5x growth
240approved Pinterest posts per month
8+ to <2hours per article with zero content-safety failures
The business need
Scale production and growth without scaling risk.
Liz needed a way to increase content output, audience growth, and marketing capacity as a sole operator. The goal was not automation for its own sake. It was a safe system that could operate, learn, and create measurable value.
Content production took too long
Audience growth required more operating capacity
Marketing workflows were highly manual
One operator needed to scale safely
The operating model
Seven agents, one Scrum system, one accountable operator.
The Scrum Master coordinates priorities. Specialists produce channel-specific work. The analyst measures results, and Liz reviews and approves external action.
01Scrum Master
02Content Writer
03Facebook Specialist
04Email Specialist
05Pinterest Specialist
06Revenue Specialist
07Analyst
Operating evidence
The agent team runs a real daily standup.
The system
Agents, applications, and controls work as one operating model.
AI-powered content engine
Specialized agents support content planning and production while Liz retains review and publishing accountability.
Pinterest application
A custom workflow handles content calendar generation, AI image and copy production, plus manual and API publishing.
Human approval system
Protected actions stop for review so speed never removes accountability from external or irreversible decisions.
Governance by design
Protected actions always return to a human decision.
01Spending money or initiating paid actions
02First-contact external outreach
03Deleting content or data
04Launching new formats or workflows
The hardest problem
AI output had to remain reliable as production scaled.
Reliability came from the operating system around the models, not from trusting a single response. Roles, context, ceremonies, controls, and performance data reinforced one another.
- Specialized agent roles and instructions
- Shared memory and repeatable context
- Scrum-based coordination and prioritization
- Human review and approval gates
- Analytics feedback and continuous iteration
The outcome
A sole operator increased capacity while preserving control.
01
Content became faster
Article production decreased from more than eight hours to under two hours.
02
Distribution scaled
The Pinterest workflow reached 240 approved posts per month.
03
Audience value grew
Traffic reached 69K peak monthly pageviews and 2.5x growth, with zero content-safety failures.
Leadership lesson
“AI needs an operating system, not isolated tools.”
Coordinated roles, shared memory, Scrum routines, analytics, governance, and explicit human accountability turn generative AI capabilities into a system that can create dependable business value.
Explore the work