TL;DR
- The bottleneck isn't writing code — it's understanding what's being built
- Specs first, frameworks second, explain-your-code third
- AI widens the gap between "what works" and "what your team actually understands"
- If a developer can't explain every decision, it doesn't ship
- Consider a fractional senior engineer who builds with AI daily
What I've Learned
The real risk isn't speed — it's the widening gap between what gets built and what your team understands.
AI makes it trivially easy for juniors to ship production code without the architectural judgment to evaluate it. That gap compounds into technical debt, hidden bugs, and a codebase nobody can debug under pressure.
Spec-driven development forces clarity before writing code. Write what the system should do in plain language before generating anything. This isn't bureaucracy — it's a forcing function.
If a junior can't articulate the behavior, they can't evaluate whether the AI's output is correct. Let alone maintain it.
Opinionated frameworks (Next.js, T3 Stack) prevent AI from inventing architecture. When AI has too much freedom, it makes bespoke, unrepeatable decisions. Constraints channel the model into known patterns your whole team can understand, review, and extend.
The explain-your-code rule is non-negotiable. If a developer can't walk you through every choice — imports, structure, variable naming, assumptions — the code doesn't ship.
Code review becomes less about "does it work" and more about "do you understand it well enough to own it."
This builds competence quickly.
One more thing: consider hiring a fractional senior engineer who builds with AI daily. You're looking for someone who can review your team's workflows — how you're generating code, validating it, testing it, integrating AI tools into your process. Their value is in reviewing how AI fits into your team, not writing code themselves.
Timestamps
- 0:00 — Jason's question about managing junior devs with AI
- 1:00 — Spec-driven development explained
- 1:45 — Opinionated frameworks and type safety
- 3:00 — Don't let AI invent architecture
- 3:20 — Testing and validation as defensive driving
- 4:00 — Three practices that force understanding
Tools & Resources
- Next.js — opinionated React framework with predictable patterns
- T3 Stack — full-stack TypeScript starter with strong conventions
- Spec-driven development — write behavior in plain language before coding
- GitHub's Spec-Kit — good starting point
- Explain-your-code prompts — e.g., "Walk through every assumption this code makes"
- Every's compounding-engineering plugin — examples
- AI code-review checklist — a structured review for AI-generated output
- Every's reviewer agents — good reference
- Fractional senior engineer — someone who builds with AI daily to review your workflows and tooling