The question I get asked most often by technical founders and CTOs right now is some variation of: "How much of this is AI going to replace?" Usually the "this" is cloud operations — monitoring, incident response, reliability work, SRE. It's a fair question and I want to give it a straight answer rather than a hedge.

My view: cloud operations is going to become largely autonomous over the next five to ten years. Not in a way that removes engineers from the picture, but in a way that fundamentally changes what engineers spend their time on. And the organisations that will fare best in that transition are the ones building the foundations for it now.

What autonomy in cloud operations actually means

Autonomy doesn't mean "the platform manages itself with nobody responsible." It means routine operational tasks — anomaly detection, alert triage, incident classification, common remediation actions, cost optimisation recommendations, configuration drift correction — can be executed with minimal human intervention at a higher speed and lower error rate than human-led operations.

The analogy I find most useful: it's similar to what happened with trading floors. Algorithmic trading didn't eliminate traders — it changed what traders do. The decisions that require contextual judgement, business understanding and novel problem-solving are still human decisions. The high-frequency pattern-matching work that previously required human attention is now automated. The total size of the market is larger, the speed is higher, and the humans who remain are working on harder problems.

What the foundations look like

Autonomy at scale requires three things that many organisations don't have yet.

Deep, consistent observability. Autonomous systems can only act on what they can see. The quality of the observability data — coverage, cardinality, consistency of naming and tagging, trace completeness — directly determines the ceiling on automation quality. Datadog as a unified observability platform, operated properly, is one of the strongest foundations available. Patchy monitoring running across six different tools is not.

Well-maintained runbooks and automation. The first wave of operational automation is runbook-based: known failure modes, documented remediation steps, automation that can execute them reliably. The organisations that have invested in runbook quality and infrastructure-as-code will be much better positioned to automate their first and second response than those whose tribal knowledge lives in people's heads.

Clear human-in-the-loop boundaries. Autonomous operations is not the same as unaccountable operations. The highest-risk actions — production database changes, cross-region failovers, security policy modifications — still require human review and sign-off. Defining those boundaries clearly, and building systems that respect them, is as important as the automation itself. The organisations that get this wrong don't lose productivity — they lose trust in the system and revert to manual everything after the first serious mistake.

What engineers do in an autonomous-first model

The SRE role doesn't disappear — it evolves. The work shifts toward designing the automation, reviewing what the autonomous systems did, handling the novel incidents the automation couldn't classify, improving the underlying system quality that the automation depends on, and making architectural decisions that affect reliability at a level the automation can't evaluate.

This is more interesting work, not less. The engineers I know who have moved into environments with more operational automation mostly find it liberating rather than threatening — because they're no longer doing the pattern-matching tasks that don't require judgement, and they're spending more time on the problems that do.

Where we are now

Within Critical Support, we're in the early stages of this transition. AI-assisted triage is reducing the cognitive load on on-call engineers. Automated anomaly detection is catching issues faster. Recommendation engines are flagging cost and reliability improvements. These are productivity tools today, stepping stones toward more autonomous operations tomorrow.

The organisations that will manage this transition best are the ones treating it as a deliberate engineering programme: invest in observability quality now, build the runbook and automation foundations now, define the human-in-the-loop boundaries now. The technology will continue to improve faster than most people expect. The operational foundations are the rate-limiting factor.

Cloud operations is going to be largely autonomous. The question isn't whether — it's whether your foundation is ready when it arrives.