Solo Lead Playbook: Running Product, Architecture and Backend with AI Agents
How one founder runs product, architecture, and backend engineering solo using a single AI coding agent, one collaborator, and a five-phase workflow.
A founder describes a concrete operating model for owning product definition, system architecture, and backend engineering with only one human collaborator (a frontend developer) and an AI coding agent handling the rest. The core claim: collapsing a team down removes the coordination latency between people, but not the judgment load — every product call, architectural tradeoff, and correctness review still has to happen, just faster and back-to-back.
The model is a five-phase loop: product framing first (treating AI output as a draft to be cut down, not a spec), then locking a database schema and OpenAPI contract before any code is written, agentic implementation against that locked contract with tests written alongside it, a parallel frontend handoff using a mock server generated from the same spec, and a final adversarial review pass before merge. For multi-tenant isolation, the piece argues for a single, consistently applied mechanism — an indexed tenant_id derived from an authenticated token, never trusted from the client — rather than mixing approaches.
The sharpest warning concerns AI-generated code that looks like it respects an invariant (soft-delete, tenant scoping, atomicity) without actually doing so — bugs that pass compilation and shallow unit tests but surface quietly in production. Mitigations proposed: small, single-responsibility files that force decomposition, and tests written alongside implementation rather than after, paired with a genuinely adversarial human review before code merges.