CASE STUDY
A 2-person farm in Upper Hutt processing 4,400+ orders a quarter, publishing to 3,870 newsletter subscribers, running events, filing board reports, and measuring environmental impact — with AI handling the operations so the team can farm.
50+
MCP tools built
4,400+
orders / quarter
3,870
newsletter subscribers
6 mo
to full autonomous stack
THE PROBLEM
Mangaroa Farms is a regenerative farm in the Mangaroa Valley, Upper Hutt. The team is small — two to three people doing everything from animal husbandry and butchery to e-commerce, marketing, events, community engagement, and board governance. The farm sells pasture-raised meat through a Shopify store, hosts on-farm events, publishes a regular newsletter to thousands of subscribers, reports to a board of trustees, and tracks environmental restoration across the property.
Before AI, every one of those functions was manual. Social media meant opening Instagram, writing a caption, finding a photo, and posting — every time. The newsletter meant sitting down with Klaviyo, writing the whole thing from scratch, building the template, scheduling the send. Board reports meant pulling data from Shopify, Klaviyo, Google Analytics, Airtable, and the impact measurement systems, then compiling it into a document. Voice notes from farm walks turned into sticky notes or nothing at all.
The result was predictable. Marketing happened when there was time, which was rarely. The newsletter went out sporadically. Social media was inconsistent. Board reports took days to compile. And the farm team — the people who should have been farming — spent half their time on admin, data entry, and tool-switching.
The farm didn't need a marketing agency. It needed systems that could run independently, connected to the tools already in use, operated by the people already on the ground. It needed infrastructure, not headcount.
WHAT WAS BUILT
Every tool was built to solve a specific bottleneck the farm was hitting. Nothing was speculative. If the team was losing time to a manual process, that process got automated.
A custom MCP (Model Context Protocol) server with 50+ tools connecting Shopify, Klaviyo, Airtable, Google Workspace, and more. This is the brain — every other system talks through it. Tasks, projects, strategy logs, and a living knowledge base (the "Open Brain") that captures every decision, meeting note, and insight. When someone asks "what did we decide about X?", the brain answers in seconds.
A Telegram bot running on a Mac Mini with Apple Silicon. Billy sends a voice note while walking the farm — the bot transcribes it locally using MLX Whisper (large-v3), then an AI agent extracts tasks, updates project records, logs strategy notes, and ingests insights to the brain. Voice notes with a #meeting tag get formatted into structured meeting notes with action items and saved to Google Drive. The farm's institutional knowledge grows from voice alone.
Instagram and Facebook posts generated and scheduled automatically. The pipeline pulls from the farm's content bank (every historical post scraped and stored), applies the farm's brand voice (built from actual content, not guidelines), generates posts, and schedules them through Buffer. Social media runs 24/7 without anyone logging into Instagram.
Two agent systems that work around the clock. Dayshift runs during NZ business hours, picking up tasks from an Airtable queue — each task triaged by AI at 7am into "AI can handle" or "needs Billy." Nightwatch handles overnight work. Both write back their results, update task status, and log what they did. The live dashboard at buildwithbilly.ai shows tasks moving through the pipeline in real time.
Nine MCP tools manage the full newsletter lifecycle. Throughout the month, items are captured to a newsletter wiki ("newsletter: lamb boxes are back" becomes an entry under Shop & Products). When it's time to send, the AI reads the wiki and brand voice, drafts the newsletter, creates a Klaviyo template, builds a campaign, and stages it for review. Billy approves, it sends to 3,870 subscribers. The whole flow from capture to send is handled through natural language.
One command pulls data from six sources — Shopify (orders, revenue, top products), Klaviyo (subscriber growth, open rates, campaign performance), Meta (social reach and engagement), Google Search Console (organic traffic), Airtable (project status, completed tasks), and the brain (strategic context and decisions). It compiles everything into a Google Doc in the board reports folder, writes a data snapshot to Airtable, and hands Billy a link to review before the board meeting.
Environmental impact tracked through two systems. Hypercerts (on-chain impact certificates) record restoration activities — planting, fencing, pest control. TrapNZ integration pulls real predator trapping data from the farm's trap lines. An evidence vault (encrypted SQLite, 660+ records) stores verified impact data with AES-256-GCM encryption. All of it feeds into the impact dashboard and board reports.
A custom marketing dashboard at dashboard.mangaroa.org pulling live data from Shopify, Klaviyo, Meta, Google Search Console, Google Business Profile, and LinkedIn. Four themed sections, 90-day windows, paginated order views. The farm's commercial performance visible at a glance without logging into five different platforms.
TECH STACK
The entire stack runs on tools a small business can afford, connected by custom integrations. Most of the infrastructure costs nothing beyond the services the farm was already paying for.
Claude Code, MCP servers, Anthropic API
n8n (self-hosted on Railway), launchd, cron
Shopify (store + checkout + Shop Pay)
Klaviyo (3,870 subs), Resend (transactional)
Airtable (projects, tasks, strategy, CRM)
Vercel, Supabase, Mac Mini M4 Pro (local AI)
Buffer, Instagram API, Facebook API
Google Workspace (Drive, Docs, Calendar, Gmail)
Hypercerts, TrapNZ API, encrypted evidence vault
MLX Whisper (large-v3), Telegram Bot API
RESULTS
Marketing runs 24/7 autonomously — social media posts, newsletter sends, content scheduling — without anyone logging into a platform.
Board reports that took days to compile now auto-generate from six live data sources in minutes. Billy reviews and sends — the compilation is instant.
Voice notes from farm walks become structured tasks, project updates, strategy logs, and meeting notes. Nothing gets lost between the paddock and the desk.
The newsletter pipeline handles 3,870 subscribers end-to-end — from item capture throughout the month to final Klaviyo campaign review and send.
Autonomous agents (Dayshift and Nightwatch) complete tasks around the clock, writing back results and updating status. The dashboard shows progress in real time.
Impact measurement is continuous and verifiable. Predator trapping data, restoration activities, and environmental outcomes are captured, encrypted, and reportable.
The farm team's time shifted from admin and tool-switching to the work that actually matters — animal welfare, land stewardship, and community connection.
TIMELINE
Built the core MCP server connecting Shopify, Airtable, and Google Workspace. Established the Open Brain knowledge base and started ingesting farm decisions, meeting notes, and project context.
Deployed the voice transcription bot on a Mac Mini. Voice notes became structured data — tasks, strategy logs, brain entries. Airtable task management with project linking went live.
Built the signal media pipeline for autonomous social posting. Added the Klaviyo newsletter pipeline — nine MCP tools covering the full lifecycle from capture to send.
Launched the Shopify analytics dashboard. Built the board report generator pulling from six data sources. Added impact measurement with hypercerts and TrapNZ integration.
Deployed Dayshift (daytime) and Nightwatch (overnight) autonomous agents. Daily task triage at 7am NZ time. Tasks auto-classified as AI-handleable or needs-Billy.
50+ MCP tools live. Evidence vault encrypted and backed up. Dashboard, agents, voice pipeline, newsletter, social — all running. The farm operates on AI infrastructure.
The same approach that runs Mangaroa Farms — connected tools, autonomous agents, voice-to-task pipelines — built for your operation. Start with a discovery call.
See also: Open Brain · AI Automation NZ