AI in Operations: How Is This Actually Supposed to Work?

As we head into 2026, AI seems to be everywhere. Every product demo, roadmap, and strategy deck seems to promise autonomous systems and self healing infrastructure.

From the outside, that sounds exciting.

From inside operations, it feels very different.

I have spent most of my career in roles that sit directly on top of a company’s infrastructure. SRE, DevOps, platform, operations, service management. The teams that carry the pager. The teams that get pulled in when something breaks and the answers matter.

Infrastructure is not abstract. It is sensitive, stateful and deeply interconnected. Small changes can have very big consequences. If you have ever been on call at 2AM, you know there is no room for “good enough”.

Those experiences are shaping how I am thinking about AI in operations.

AI will change operations, but not through autonomy first

I do believe AI will change how we operate systems. I just do not believe 2026 is the year we hand over the keys.

Most environments are messy. Hybrid. Full of legacy decisions. Data is noisy and context is often missing. In that reality, humans are going to stay in the loop for a long time.

Not because AI isnt good enough, but because accountability does not disappear just because a model is involved.

The question that keeps coming up

Whenever I hear about AI agents making changes in production, I keep coming back to one question:

How is this actually going to work? How does this integrate into existing workflows? How is access controlled? How is intent validated before action? How is risk understood before something changes?

Most organisations already struggle to maintain a clean audit trail of their human actors. Who has access, who made a change, why it was made, and what it impacted.

That problem exists today, without AI. So before we introduce AI actors into environments, there are some questions i keep coming back to:

How do we differentiate AI actors from human actors? How are AI actions logged, attributed, and reviewed? How do we ensure automation stays within well defined guardrails? If you cannot solve for this within your infrastructure, autonomy will becomes a liability, not an advantage.

Operations is judgement

Operations is not just execution. It is judgement.

Judgement about risk tolerance, business impact, timing, and the trade offs between speed and safety.

That judgement should not live only in people’s heads. It needs to be captured in runbooks, playbooks, and service management practices that reflect how the organisation actually operates under pressure. I have worked in environments with good runbooks, clear escalation paths, and strong service management, and there were still moments where the documented “right” action was not obviously the right thing to do in that moment.

Runbooks cannot anticipate every variable. They cannot fully account for shifting business pressure, partial failures, incomplete data, or the way multiple systems interact under stress.

They give teams a baseline. They reduce uncertainty. They narrow the decision space. But they do not remove the need for human judgement.

That is why I am cautious about the idea that AI can simply follow runbooks and make the right call. At best, AI can help interpret what is happening and suggest options. The final decision still belongs with people who understand the risk, the impact, and who will be accountable for the outcome.

Where i see AI actually helping

With strong runbooks and service management in place, AI becomes genuinely useful, not as a decision maker, but as support for human judgement.

It can pull together signals faster than any person, correlate what is happening across systems, and suggest likely causes or next steps. AI can recommend. AI can correlate. AI can simulate.

But humans still decide. They decide when to act, how far to go, and what risk is acceptable in that moment. Without clear ownership and audit, AI just accelerates reaction. With them, it supports better judgement.

That is also how I think about OpsZ. We are not building autonomous infrastructure. We are focused on making it clear who or what is acting, why an action was taken, and whether it stayed within agreed guardrails.

Differentiating AI actors from human actors. Making actions auditable and reviewable. Putting boundaries around what automation can and cannot do.

In my experience, those basics matter far more than autonomy ever will.

What I actually expect to see in 2026

I do not expect to see fully autonomous infrastructure everywhere.

What I do expect is teams doing harder, less visible work:

• Cleaning up access and audit gaps

• Defining clearer ownership models

• Writing better runbooks and playbooks

• Designing workflows where humans and AI work together safely

AI has a way of exposing weak process and unclear accountability very quickly.