Post

In 2026, manually ssh’ing to a production server and changing its configuration is a clear anti-pattern. Let’s do the same for software changes. Humans don’t change software. Only their agents do. Sounds irresponsible until you examine the ramifications of that constraint.

aicoding.leaflet.pubThe Conversation Is the Commit

2 reposts6 likes

Doesnt this just move the convention problem though? If my prompt was just "increase retry timeout by 5 seconds" the conversation is no more enlightening than the code change.

Agreed though doing this on a team in public would I think make people less likely to do this. There would also be conversational context around that message. I would probably also have agents in channel whose job is to ask eg “why?” When something like that happens

Are coding agents in groupchats a thing already? And if so, do they work close to as well as silo’ing design team conversations would? Not even for a purist no-AI result. Even using agents to process Slack chats into documentation sounds like spending fewer tokens for better results