How I Build with AI
Three workflows for three contexts. The tools change depending on who I'm building for, but the principle is the same: AI handles the execution, I handle the judgment.
When: For clients who know roughly what they want but need help articulating it clearly and building it right.
Discovery Call
Recorded session with Fireflies note-taker. PM-focused questions: who are we building for, what pain points are we solving, what do they like and dislike, what does success look like.
Client Homework
Client identifies competitors they like/dislike and why, defines their goals, mission, and purpose. This surfaces the real priorities before any code gets written.
Build the Brain
Compile everything into NotebookLM as the central knowledge repo. Set up Notion as the single source of truth — product docs, strategy, branding, design specs — connected to Claude Code via MCP.
Collect Assets & Review
Gather images, fonts, branding assets, and design files. Drop everything into the project folder. Read through all generated docs to make sure everything aligns with the client's vision. This checkpoint usually takes a few days.
Connect the Stack
Wire up MCPs based on the project: Figma for design teams, Supabase for backend, GitHub + Vercel for deployment, PostHog for analytics. Each project gets exactly the integrations it needs.
Build
Claude Code builds the full product against all the artifacts — docs, designs, assets, and live integrations. Notion tracks tasks in real time so I can watch agents run, see what they're working on, and catch issues as they happen.
Production Output
Example: Swob — built 5 prototypes as investor demos and living specs for the founding team.
Every product on this site was built using one of these workflows. The common thread: AI is the force multiplier, but product judgment — knowing what to build, what to cut, and when to change direction — is the skill that makes it work.