How AI Handles Post-Launch Software Maintenance
Can AI maintain a software project after launch? In practice, yes for a lot of the routine work: watching production, grouping errors, preparing dependency updates, and drafting patches. No for the pa
#software-development#software-engineering#software-architecture#technology#artificial-intelligence#ai-agents#ai-tools#web-development#webdev
Responses
Markdown supported · ⌘↵ to post
The rule that every task ends with a person making a decision is the right boundary, and it is clearer than most attempts to sort maintenance into safe and unsafe work.
Treating triage as a hypothesis that cites log lines is the detail I liked most. A summary that says the payment job is failing is easy to accept without checking. One that points to the exact lines lets the reviewer agree or disagree in a minute.
The maintainer artifacts matter just as much. A written spec, a diagram and a change record are what make AI assistance useful months after launch, because the model, like a new developer, can only reason about what has been written down. On client handovers we treat those three documents as part of the delivery, not as optional extras.
The weekly and monthly cadence is sensible. How do you decide when a clustered set of tickets becomes a project rather than maintenance?