NoOps: Will AI Eliminate Traditional DevOps?

In response to a LinkedIn post I came across (If you are part of the AWS Certification & Training).

I hit the character limit 😅, so extending on that here.

DevOps Automation

DevOps has been trying to automate itself out of its role since its inception. That was the whole point of the role in the first place and yet still many orgs are facing:

• Totally custom setups and never-ending edge cases.
• Operational dysfunction.
• Multiple disjointed tools that perform the same functions.
• Development teams operating in silos that don't conform to the federated guidance DevOps is attempting to implement.
• Incomplete monitoring systems or multiple monitoring tools that don't cross communicate.
• Incredibly ever complex logs with vague descriptions as to what actually went wrong.
• Office politics where confusion exists around role and responsibility when the pipeline breaks.

Current Challenges

Many orgs are still:

• In the phase of getting away from monolithic apps into microservices.
• Migrating deployments to the latest trendy tool instead of sticking to what works and building on that.
• Depending on a single DevOps engineer or a very small team to handle operations optimization, monitoring, analytics, finance/cost impact reporting, deployments, pipeline optimization, and most recently implementing AI pipelines.
• Still scaling to implementing a data plane, control plane, UX plane and control-tower setup as opposed to the per-client microservices architecture where IaC (Infrastructure as Code) is copy-paste into a new customer environment instead of building on a central product.

The Practical "Promise Land"

What LLM's can do for DevOps is:

• Improve testing.
• Slightly optimize log analytics.
• Streamline cross team communications.
• Optimize planning and architecture.
• Complete and optimize documentation.
• Shortcut anything that involves large blobs of text.

What's Still Relevant?

Skills that are still very much needed include:

• Operational discernment.
• Architecture and planning.
• Knowledge and experience of the tools you are implementing with AI guidance.
• Eye for detail.
• Effective decision making and executive function.

Forward Facing Guidance

What we fail to communicate effectively in this current cycle of "AI" is that the majority of the conversations are actually around LLMs or Large Language Models. These are a tool designed to help with ✨language✨. Let's apply the tool where it makes sense!

Intelligence is measured under a plethora of layers and dimensions. Language models are NOT going to solve all our problems.

Legacy Systems Still Valued

In a world where Fax machines, COBOL, Fortran, DB2 and Mainframe systems are still commonplace and VERY heavily relied upon, I'm highly skeptical that all of these are going to migrate to the current cycle of "AI" (ahem, read: LLM's) and these so-called "NoOps" systems overnight.

It's going to take a great deal of planning, architecture, testing, building and more office politics before something like this is commonplace -- even if it does solve all our problems tomorrow (which, I'm also skeptical it will).

Conclusion

So, rest assured, fellow DevOps and CloudOps engineers: Your job isn't going anywhere. It's just another tool in the box of chaos we deal with on a daily basis!