Start Somewhere. Building Your Own Products with AI

Christine Vallaure opens by asking the room not to feel inspired. She was here a year earlier, got lovely emails afterwards, and wants something else this time: build. Her platform Moonlearning started for designers and now teaches people to build with AI. Her message: you don't need to code, you need your core skill plus AI literacy.

What the talk covers

• The job market in two pictures: AI eats the junior layer from the bottom up, while the middle, where most freelancers sit, stays and gets a new layer on top. Her metaphor is the hat. Your skill is the core, the model is the brain, and tools like ChatGPT or Lovable are hats you put on it. Every new tool runs on the same few models, so understanding models cuts the overwhelm.
• Three ways to build, with demos:
  • Vibe coding (Lovable, Bolt, Replit): describe what you want and it builds front end, back end and database. Her examples: a brief generator that turns messy client emails into a list of deliverables, an Alpine crossing planner and a finance tracker that categorises a pasted bank statement. Good for prototypes and in-house tools, not for complex products or sensitive data.
  • Pair coding (Cursor, Claude Code): you do more of the product work and learn how code is structured. Treat it like an intern: would you ship it without a review?
  • Agentic AI: a hat with hands. She runs a small team of agents in Claude Cowork (a lead, a business agent, a writer in her voice, a designer), defined by a plain markdown file. MCP connects it to tools like a USB-C cable, for example her Figma designs.
• Examples: a weekly newsletter pipeline with a researcher and a writer, and a psychology professor who built synthetic patients for therapy students within minutes.
• The honest part: it fails, hallucinates and burns tokens. It's the wild west, so you have to tinker.

Who it's for

Freelancers who think building with AI is only for developers.