What twenty years of DevOps has failed to do

AI, observability & autonomous changes

• Some predict LLM-based “super-agents” will commoditize observability vendors by cloning features cheaply, at least for simpler integrations.
• Others argue observability/ops is highly bespoke, full of version-compatibility landmines and snowflake systems, making it one of the hardest domains for agents to automate.
• Several commenters report mixed real-world results: AI occasionally finds subtle bugs or does strong code reviews, but also produces wrong “fixes” and nonsense root causes. Trust is fragile, especially after bad vendor demos.
• There’s skepticism about chat-based interfaces to dashboards: if devs ignored dashboards before, they may ignore chat too, and LLM answers are not reliably trustworthy.

Accountability for production failures

• One camp says fully autonomous production changes are obviously a bad idea; each change must have a human owner who understands and stands behind it.
• Others note humans already routinely deal with legacy or absent authors, so “code you didn’t write” is normal.
• Some expect leadership to tolerate outages and invest in better testing/mitigation rather than abandoning autonomous changes.
• A cynical view: organizations may blame the LLM and “prompting” rather than accept human responsibility.

What “DevOps” means & whether it failed

• Definitions vary wildly: methodology, role, rebranded sysadmin, collaboration pattern, or just “owns Jenkins and k8s.” This semantic overload is seen as a core failure of the “movement.”
• Several argue DevOps-as-practice (tight dev–ops collaboration, automation, shared ownership) works well; DevOps-as-title or cost-cutting strategy is what failed.
• Some say DevOps is effectively “dead” or a “zombie,” kept alive by vendors and HR as a buzzword.

Dev vs Ops: skills, silos, and org design

• Many emphasize dev and ops are distinct disciplines; expecting one person or team to master both at scale is unrealistic.
• Others stress the goal should be shared mental models and close collaboration, not collapsing roles into “interchangeable EngDocs/DevPM/DevOps.”
• Management choices loom large: underinvesting in ops, creating DevOps bottlenecks, or using DevOps to shift responsibilities without authority are framed as organizational, not technical, failures.

Tooling, Kubernetes & configuration pain

• k8s, Terraform, and similar tools are criticized as over-complex, ill-matched to certain workloads, and often used without sufficient expertise.
• YAML is widely disliked as a core “DevOps failure”; people advocate treating it as a wire format and generating it from higher-level languages or newer config systems like CUE.