Design and planning become the bottleneck when agents write all the code¶
Insight: When coding agents handle implementation at scale, design and planning become the primary bottleneck. Steve Yegge's Gas Town agent orchestrator runs dozens of coding agents simultaneously, revealing emergent patterns: agents with specialized roles under hierarchical supervision, persistent roles with ephemeral sessions, continuous work streams rather than batch tasks, and merge queues with agent-managed conflicts. The price (thousands of dollars monthly in API costs) is extremely high but the potential value — near-autonomous software development — is unprecedented.
Detail: Appleton identifies key questions about when to stop reviewing agent-generated code, organized by: domain and programming language (well-specified domains work better), access to feedback loops and definitions of success, risk tolerance, greenfield vs brownfield projects, number of collaborators, and experience level. The Gas Town model represents the extreme end of vibe coding at scale — fully autonomous agent swarms — and forces the industry to reckon with whether human code review remains necessary or becomes counterproductive overhead.