← All work

Artspace — Arts Consortium Platform

Artspace is a full-stack platform connecting freelance artists into consortiums, built entirely via conversational AI development as the central PhD case study.

Artspace platform interface showing the consortium-building dashboard for freelance artists with project listings and team management
Artspace platform interface showing the consortium-building dashboard for freelance artists with project listings and team management
Problem

Freelance artists struggle to compete for large commissions individually. Could a platform help them form temporary consortiums to bid on bigger opportunities?

Results
  • Full-stack application built entirely through vibe-coding
  • Central case study in PhD research documenting the 80:20 pattern
  • Revealed key failure modes in AI-assisted development

Artspace emerged from the same impulse as several of my Brighton-focused projects: how do you sustain creative community infrastructure when physical spaces keep closing?

The Challenge

As part of an Arts Council England challenge during my CDT programme, I set out to build a consortium-builder for freelance artists. The rules I gave myself were deliberately constrained: only conversational prompting, no manual code fixing. I wanted to see what happened when you try to build a real platform purely through vibe-coding.

What Went Right

Artspace became a full-stack platform with role-based access control, document management, AI-assisted guidance for grant applications, and a commissioning workflow. The speed of iteration was genuinely impressive. Features that would have been entire sprints in a traditional project appeared within sessions. The AI could scaffold React components, set up API routes, configure database schemas, and wire everything together through conversation.

What Went Sideways

Two failure modes became especially visible. The first I call the scaffold problem: the AI builds beautiful facades with non-functional internals. Everything looks wired up. The UI is polished. But permissions don’t cascade correctly, edge cases aren’t handled, and data that should be private isn’t. It’s convincing enough to demo but not robust enough to trust.

The second is infrastructure blindness. The AI generates code that assumes hosting environments, DNS configuration, SSL certificates, and database connections will just work. It has no model of the gap between “this runs on my machine” and “this runs on the internet for real people.”

Why It Matters

Artspace became one of the two central case studies in my PhD research. Not because it failed — it works, and the core idea is sound — but because it illustrates exactly where the boundary sits between what AI-assisted development can do and what it can’t. The things that broke weren’t bugs in the traditional sense. They were gaps in understanding that no amount of prompting could bridge. Configuration, permissions, data boundaries, security. The invisible infrastructure that working software depends on. I wrote about Artspace and this 80:20 pattern in What Happens When You Let AI Write All Your Code.