Ask HN: Did agentic coding change the way you think about commit granularity?
Jujutsu is trending on the homepage, and the topic is using discipline when dealing with version control. Six months into working agentially on a daily basis, something changed for me.
When each coding session has one specific aim to achieve before the agent starts: one migration, one service, one abstraction layer, the diff itself is already the unit you can review. The whole commit granularity issue went away. No more interactive rebases for me.
Conversely, coding sessions with a large enough scope, implementing dark mode, and fixing the authentication flow across three subsystems will always end up with a diff that is really hard to read, regardless of commit granularity. Inconclusive diff is the result of bad scope, not bad commits.
When each coding session has one specific aim to achieve before the agent starts: one migration, one service, one abstraction layer, the diff itself is already the unit you can review. The whole commit granularity issue went away. No more interactive rebases for me.
Conversely, coding sessions with a large enough scope, implementing dark mode, and fixing the authentication flow across three subsystems will always end up with a diff that is really hard to read, regardless of commit granularity. Inconclusive diff is the result of bad scope, not bad commits.