Ask HN: How do you maintain flow when vibe coding?

Article

• HN thread on managing focus and flow while running parallel AI coding agents
• Core tension: parallel agents create serial bottlenecks at the human review layer
• Strategies discussed: worktrees, single-project focus, devlogs, task-by-task discipline

Discussion

• Consensus: running 3+ agents means you’re the bottleneck — effective output ≈ 1 agent
• Several commenters abandoned agentic mode entirely due to accumulated debt and lost awareness
• Practical tips: git worktrees, /devlog skills, “do the dishes” as third async stream
• Contrarian: one commenter says Opus still outputs wrong code too often to parallelize

Discuss on HN