Limit parallel sessions, accept idleness
Many cap themselves at one to three agent sessions and treat waiting time as a break, arguing context switching is the real cost and human attention hasn't changed.
Review is the true bottleneck
Parallel agents produce huge PRs that still need careful human review, so going serial may actually land code faster; faster models help but tests and CI still add waits.
Post reads as AI slop
A large contingent called out Claude-style phrasing in the blog post and found it ironic for an article about AI brain fry; the author shrugged that the idea was captured fine.