AI Incubator for Solopreneurs | Session 1: Vibe to Live App
Session
AI Incubator with SixFactors, Aug 14, 2026, Fabrik, Dumbo NYC
My role
I attended as a mentor and participant, helping people troubleshoot as they built, documenting where they got stuck, and demonstrating parts of my own workflow to the group.
What we were trying to accomplish
The goal was to move beyond chatting with AI and start building an actual application using Claude Code, Terminal, GitHub, Vercel and Supabase.
What I noticed
The technical material wasn’t necessarily the biggest obstacle. The missing foundations were.
People were being introduced to agents, APIs, MCPs, SDKs and application architecture while still trying to figure out whether they’d downloaded an app correctly or what Terminal actually was.
Once someone is lost at that foundational level, every new instruction compounds the problem.
Where people got stuck
Setup came before learning. Downloading, installing, opening and authenticating the required applications took significant time.
Basic terminology wasn’t defined. Terminal, artifact, API, MCP, SDK, frontend, backend and architecture were being used before everyone had a working mental model for them.
Terminal was especially unfamiliar. People needed to understand what it is, where they’re typing commands, where projects and folders live, and what it means to “run” something before being given commands to enter.
Too many tools appeared at once. Claude, Claude Code, Terminal, GitHub, Vercel, Supabase and Visual Studio Code each had a purpose, but the relationship between them wasn’t established first.
There wasn’t enough structure around the learning journey. Participants needed an agenda, breaks, setup instructions and a simple map showing where they were in the process.
What helped
Showing the process visually made a difference. Dean connected my laptop to the main screen, which allowed the group to see what I had done rather than only hearing instructions.
Claude Artifacts also provided a useful bridge between an idea and code because participants could actually see a prototype of what they were creating before worrying about how the full application would be built.
Breaking instructions into very simple A → B → C blocks also felt much more appropriate for this audience than explaining the entire technical architecture upfront.
What I’d change next time
Send a pre-session setup guide with every download link, subscription requirement and instructions for verifying that each application is installed and working.
Start with a one-page visual map of the tool stack: what Claude does, what Terminal does, where Claude Code fits, why GitHub is used, what Supabase handles and what Vercel does.
Define unfamiliar terms when they’re first needed, rather than introducing concepts like MCPs and SDKs before participants have a reason to care about them.
Build in breaks and checkpoints so everyone can get to roughly the same place before introducing another layer.
End with a take-home guide containing the workflow, important commands, definitions, tips, common mistakes and what participants should try next.
What I’m keeping
Don’t teach the advanced thing until people have a mental model of the basic thing.
Someone doesn’t need to understand MCPs to build their first simple app. But they absolutely need to understand what Terminal is if you’re asking them to work inside it.
The sequence matters: orient → show → do → explain the deeper technical layer when it becomes relevant.
What this is teaching me about AI education
There is a big gap between making powerful AI tools accessible and actually making people feel capable of using them.
Teaching AI isn’t just simplifying technical language. It’s designing the learning experience around what the learner sees, knows and needs at that exact moment.
The instructor already knows how all the pieces connect. The learner doesn’t. Good AI education has to build that map for them, one piece at a time.