Project

Sandbox

AI-native content and curriculum engine (Sandbox Creative LLC). Three layers: a community and brand platform co-led with Daniel Fitzgibbons, a Labs client-services arm building custom AI systems for businesses, and a University tier at $49/mo. Flagship delivery: SandboxOS — 9 agents automating a 72-lesson curriculum end-to-end.

Role Co-Founder, AI Architect & Lead Developer

Stack Claude Code · MCP · Python · OpenRouter · Docker · Notion API

Status Active — Labs engagements open

Since January 2026

Three layers

Sandbox operates as three distinct but connected layers. The parent business is a community and brand platform — content, voice, social — co-led with Daniel Fitzgibbons. Sean does not touch outward-facing copy here; Daniel owns that lane entirely.

Sandbox Labs is the client-services arm: custom AI systems for businesses. The delivery model is productized — requirements discovery, architecture, build, and ongoing operations retainer. Labs translates non-technical business requirements into multi-agent architectures that run with minimal human intervention.

Sandbox University is the $49/mo tier — a structured curriculum product for individuals who want to learn how to build with AI without needing a custom Labs engagement. The curriculum engine SandboxOS was originally built to power this tier.

Flagship build: SandboxOS

SandboxOS is a production multi-agent operating system built to automate a 72-lesson curriculum end-to-end. The client needed to scale content production without scaling headcount. SandboxOS is the system that makes that possible.

Agents 9 specialized agents: Orchestrator, Copywriter, Art Director, Audio Producer, Curriculum Writer, and domain-specific specialists. Each has a narrow authority scope and declared contract.

Model routing Tiered model routing — heavy reasoning tasks route to larger models; high-volume formatting tasks route to faster, cheaper models. Cost-performance optimization by task type.

Human gates Human-in-the-loop approval gates at key production stages. The system generates; humans approve before publication. Automation without blind trust.

Command center Centralized Notion command center for operator control — status visibility, approval queues, and output archives in one interface the client already uses.

Output 28 assets per 14-day production cycle across a 72-lesson curriculum — scale that would require 3–4 human content producers to match manually.

The Labs delivery pattern

The core Labs motion is requirements translation. Clients come in with a business problem expressed in business terms. The engineering translation — decompose the workflow into discrete agent roles, route each role to the appropriate model tier, insert human gates at the right checkpoints, expose the whole system through an interface the client already knows — is the design work that determines whether a project succeeds.

The same architecture pattern is reusable across different client domains. Labs is building a playbook from these engagements, not just delivering one-off systems. Each build sharpens the next one.

Active and upcoming engagements

Current Labs pipeline targets food and hospitality businesses — a sector where AI operations (scheduling, customer-facing assistance, inventory signaling) has clear ROI and low technical prerequisite on the client side. The Chill'n ice cream shop engagement was the first intake in this vertical, informing the architecture template for that category.

University pricing is set at $49/mo. Labs custom engagements are priced per project (base $4K + $300/mo for recurring ops) with per-additional-location pricing for multi-site clients.