Josh Pigford, who sold his last startup for $4 million, is now building 5 products simultaneously as a solo founder using a structured set of AI skills inside Claude. The four skills driving his workflow: /build (a 4-step sequence covering research, planning, tracking, and implementation), /adversarial-code-review (Claude Opus generates code, GPT-4 reviews it), /but-for-real (forces the AI to surface mistakes it glossed over), and /learnings (automatically updates a CLAUDE.md file so the AI improves its own context over time).
The /learnings skill is the one worth understanding in detail. Rather than manually tuning prompts, Pigford has the AI write its own persistent instructions based on what worked and what failed. This is a compounding loop, not a one-time setup. The adversarial review pairing, pitting two frontier models against each other on the same codebase, is also a concrete workflow most developers have not systematized. These are not abstract productivity tips. They are specific, repeatable commands built into a 25-year solo builder's daily process.
The video also includes a live demo on designing websites that avoid generic AI aesthetics, and Pigford addresses how non-technical founders can enter this workflow. The interview is hosted by Peter Yang and runs just under 30 minutes. The full breakdown, including the skills themselves, is linked in the description. If you are building alone and relying on AI as a core collaborator, this is a direct look at how someone doing it at scale has structured the system.
[WATCH ON YOUTUBE →]