Managing AI coding assistants feels like being a billionaire complaining about household staff. That's what one developer wrote on Hacker News this week. They weren't wrong.
A thread on vibe coding and flow state struck a nerve.
Developers are shipping faster with AI. Their concentration is fractured.
One developer described their setup: Claude in fast mode with low thinking, paired with NVIDIA's local text-to-speech model and hot reloading for UI work. Higher thinking modes might allow more parallel work, they said, but it's exhausting. Others cited Paul Graham's "Maker's Schedule, Manager's Schedule." Flow needs uninterrupted blocks. Managing agents breaks them.
Quality drops. Overhead piles up.
Stanford research from Vaithelingam et al. backs this up. Developers using GitHub Copilot finished tasks 55.8% faster but wrote less secure code, showing "automation bias" by trusting AI output despite errors. Claude Opus 4.6 writes code with fewer bugs than he does—but he still discards half its solutions. The problem: AI lacks "taste," over-engineering solutions and producing code humans struggle to debug.
Cognitive load moves from writing syntax to reviewing semantics.
Some developers report that fixing AI hallucinations eats the time saved on drafting. The developers who've been vibe coding longest are already pulling back. Fewer agents. Simpler setups.