Should an AI coding agent ever be allowed to merge its own PR?

A discussion on reddit.com explores the boundaries of AI coding agents, questioning at what stages they should be allowed to operate autonomously. The conversation delves into whether agents should be permitted to write code, run tests, commit, open PRs, merge, or deploy. The author notes that passing tests doesn't automatically mean an agent should be allowed to ship code, highlighting a key distinction in the debate.