Connecting GrowthPod’s Operations Through Five Custom AI Agents.

The Challenge

GrowthPod already had a working internal platform coordinating high-volume creative production. The challenge was making that system more reliable as the agency scaled: safer releases, clearer handoffs, stronger quality control, and AI that could participate in real work without taking control away from the people accountable for clients and production.

The assignment was not to replace the existing Mothership application or claim the original product as our own. The Liminals improved it around the way the team actually worked, then built governed agent workflows into the gaps where repetitive research, review, handoff, and engineering work created drag.

Real screenshots from GrowthPod’s live Mothership app, composed on a solid lime background. No test records or generated interfaces.

Five Agents, Clear Boundaries

Reece supports research and knowledge operations. Cue reviews creative work against brand and brief requirements. Han manages guarded creative handoffs when work is ready. Otto turns likely product issues into isolated investigations, tested changes, and reviewable releases.

Each agent was designed around explicit permissions, observable state, recovery behaviour, and human approval. The goal was not maximum autonomy. It was useful autonomy: systems that could move work forward, explain what they had done, stop safely, and leave judgment with the team.

Improving Mothership Around Real Work

The work strengthened the application beneath the agents: isolated Preview and staging environments, protected release gates, authenticated browser testing, production checks, recovery paths, and rollback-safe delivery. More than 30 releases were verified through that process.

Inside a 62-brand production environment, the system has been tested against the operational scale that matters: 799 loaded pipeline items, 416 Primary Text rows migrated under guarded ownership rules, and 682 existing animation jobs brought under a controlled autonomy contract. Those figures describe the operating environment and completed application work—not 62 clients or revenue attributed to a single feature.

Estimated Capacity, Not Invented Savings

GrowthPod reports operating at approximately $500K–$600K in annual revenue. Across the five agents, workflow automation, and application improvements, we estimate the engagement recovered roughly $10K–$30K in annual operating capacity. That range reflects time and delivery capacity returned to the business; it is not an audited cash-savings figure or a claim that the same amount appeared directly on the bottom line.

Summary

The result is an operating system that makes AI useful to the whole business. Research, creative review, handoff, and engineering now move through specialized agents connected to the application and communication tools the team already uses—while approval, client judgment, and production control stay firmly with people.