Ghost Hat Studio
Work
Case study · Architecture & design

Backstage: agentic modernization

Backstage, a marketing platform for live-events venues and promoters, was losing users to friction. We designed the modernization. Keep the proven Rails engine, and rebuild the experience around an eight-agent team.

Role  Architecture & product design Year  2026 Stack  Rails (preserved), REST API, Next.js, Tricorn agent runtime Status  Design delivered · client building in-house
01

The problem

Backstage helped venues market their shows, but the product had grown into a wall of forms. Adding a single show meant a 20-plus-field form across seven sections; starting a post meant five levels of cascading dropdowns. There was no monitoring, so ticket-sales surprises only showed up at show time, and a calendar-first layout buried the shows that needed attention. The friction was driving clients away.

Underneath, though, was a real asset: a mature Rails application with a working social-publishing pipeline, billing, auth, and an early AI prototype. The job wasn’t to rip it out, but to make it feel modern and intelligent.

Today
20+ field show form across seven sections
5-level cascading dropdowns just to start a post
Manual template selection per platform, per phase
No monitoring, so sales surprises hit at show time
Calendar-first view buries the shows that matter
After
Three fields → the agent drafts the plan in the venue’s voice
One-tap approve, edit, or regenerate per post
Templates become an AI fallback, not the starting point
A live command center with proactive interventions
Show-first dashboard with status at a glance
02

Evolve the engine. Reinvent the experience.

The architecture protects the investment in Rails while opening the door to an agentic experience. You add a clean, versioned API to the existing app, build the new front end in Next.js, and promote the AI prototype to a real feature. The business logic, social publishing, billing, and auth all stay exactly where they are. They just get a better interface and a smarter layer on top.

A new experience on a preserved foundation
New
Next.js front endA show-centric command center: agent-drafted strategies, live monitoring, and one unified design system.
REST API · v1  ↓
New
API layerClean versioned endpoints for shows, posts, venues, strategies, and AI generation. It wraps the existing service objects without touching them.
same models · same database  ↓
Kept
Rails backend · preserved156+ service classes, Sidekiq jobs, and the Facebook / Instagram / X publishing pipeline keep running. Auth and billing stay put, so there’s no migration risk. The AI prototype becomes a real feature.
03

An eight-agent team, three pipelines

Instead of “AI generates strategy,” the whole product is reframed around a team of specialized agents on a shared runtime. They show up consistently on every screen, from onboarding (“meet your team”) to the command center. Three pipelines carry the work, each built on the same principle.

Automate the friction.
Humans inspect. Humans accept.
01
Strategy generation
H
human
Add a show in three fields
Artist, date, ticket tiers.
AI
agent
Researches the artist and drafts the full plan
Web search, fan signals, and the venue’s brand voice → a sequence of posts across the timeline.
H
human
Reviews, edits, approves
Per post: approve · edit · regenerate · delete.
system
The existing Rails pipeline takes over
Sidekiq schedules and publishes to Facebook, Instagram, and X.
02
Monitoring & intervention
system
Ticket sales stream in continuously
Integrated from the venue’s ticketing partners.
AI
agent
Evaluates against benchmarks and trends
Lagging? Surging? Off pace? Fires an alert at the threshold breach.
AI
agent
Drafts an intervention in the venue’s voice
An urgency post, a paid-boost recommendation, a contest idea.
H
human
Approves, edits, dismisses, or snoozes
One tap, and each response trains the system.
03
Voice & sensitivity refinement
H
human
Approves, edits, or dismisses every output
Every interaction is signal.
system
Captures the delta as feedback
What got accepted vs. rewritten; which alerts fired vs. got dismissed.
AI
agent
Refines the venue voice and alert sensitivity
It gets sharper per venue the more it’s used, so the product compounds in value.
04

The outcome

We delivered the modernization architecture, the agentic-pipeline design, a full set of wireframes covering every screen and flow, and a locked design system. That was the whole blueprint for the build. The client was confident enough in the plan to take it in-house and build it themselves with Claude.

The strongest endorsement of a plan is a client ready to build it themselves.
05

What it demonstrates

  • Modernize without a rewrite. A clean API and a new front end on top of the engine they already had. No migration risk, nothing thrown away.
  • Agentic UX, not a chatbot. A team of specialized agents surfaced consistently across the product, doing real work in the background.
  • A person stays in the loop everywhere. Every agent output gets approved, edited, or dismissed, and every one of those decisions trains the system.
  • Pipelines that compound. Monitoring and voice refinement make the product sharper per customer the more it’s used.
  • Architecture, design, and a plan they could build from. Detailed enough that the client shipped it on their own.
Sitting on a legacy product that should feel intelligent?
Thirty minutes. We listen, we ask questions, we find out whether there’s a fit.
Book a call