Skill Soup In Practice: Building Without Specialists

The skill soup idea: a builder who knows a little about many domains, combined with AI, outperforms a specialist in any single one. It sounds good in a blog post. The real test is shipping product with it. Three months in, the data is coming in.

The Proof

This platform has a WordPress side, a FastAPI backend, a DynamoDB store, a Next.js frontend, an SES email pipeline, and a custom admin UI. On paper that is five specialists. In practice it is one me plus AI.

Each domain gets handled when it comes up. I do not need to be world-class at React. I need to know enough React to direct Claude Code effectively, recognize when the output is wrong, and fix the edge cases. Same for SQL, for DynamoDB, for SES DKIM records.

What Breadth Means

Breadth is not shallow familiarity. It is enough depth in each domain to spot errors and make good tradeoffs. I cannot write a complex Postgres query from scratch quickly. I can read the one Claude Code wrote, spot the missing index hint, and ask for a revision.

Specialist: deep in one thing, hires out the rest
Generalist (old): shallow in many, nothing ships fast
Skill soup: functional depth in many + AI speed = solo delivery

The middle row used to be the dead zone. AI made it the winning position for solo builders.

Where It Breaks

When the problem requires true specialist judgment: performance tuning Postgres under heavy load, securing a payment flow, designing a distributed consensus protocol. For those, I hire. The skill soup covers 90% of client needs. For the other 10% you know when you need the help.

How To Build Yours

Pick ten domains you want functional depth in. Spend two weeks on each. You will not become an expert. You will become dangerous enough to direct an AI and recognize good output. That recognition ability is the whole skill soup unlock.

The thesis keeps holding up. I ship things that used to require a team. The leverage is real.