Prime is right about AI
Theo responds to ThePrimeagen's recent skeptical takes on AI coding agents. He concedes most of Prime's specific complaints (slop, junior devs not learning, review fatigue) while pushing back on the framing that agents are net-negative for serious engineering.
Where Prime is right
- Vibe-coded slop PRs are a real and growing review burden.
- Juniors who let the agent do everything stop building intuition.
- Tooling that auto-merges agent output without a human gate is harmful at any scale.
Where Theo pushes back
- The productivity ceiling for experienced engineers using agents as power tools (Claude Code, Codex CLI) is genuinely higher; the problem is process, not the technology.
- "AI replaces juniors" framing is wrong — what's collapsing is the market for writing boilerplate, not the market for engineers who understand systems.