The runbook maintenance problem is real, but the harder challenge is trust.
I've covered DevOps and SRE practices for years. The worst runbook isn't the missing one—it's the one that's 80% right but catastrophically wrong in one step. During a production incident, you follow it exactly, and that 20% breaks something worse.
AI-generated playbooks face the same trust gap. They can be stack-aware and well-structured, but the on-call engineer at 2am doesn't know if this playbook has ever been validated in their actual environment. Generic templates at least fail obviously. AI-generated ones can fail confidently.
The breakthrough would be tying runbook generation to actual incident resolution—capture what the team did to fix it, then generate the playbook from that session. Not upfront templates, but post-incident extraction. That way the runbook is already battle-tested before it gets used again.
The problem you're solving matters. Just make sure the solution doesn't trade "no documentation" for "documentation we can't trust."
Good luck with the build.