AI Will Replace Half of DevOps Jobs by 2028 — And That's Fine

Let me say what most DevOps practitioners are thinking but won't say at conference panels: a significant portion of what we call "DevOps work" today will be fully automated within three years. And the sooner we reckon with that honestly, the better positioned we'll be.

This isn't a hot take designed to generate LinkedIn outrage. It's a conclusion I've reached after watching agentic AI systems operate in production environments and seeing what they can already do — quietly, reliably, and at a cost far below a senior engineer's salary.

Nicole Forsgren, co-author of Accelerate and the researcher behind the DORA metrics that became the industry benchmark for DevOps performance, has noted that the next wave of DevOps evolution won't be about faster pipelines — it'll be about intelligent systems that can reason about their own performance and adapt. The data is pointing in one direction. The question is whether the people in the field are paying attention.

---

What's Already Being Automated

I've watched these things happen this year:

Infrastructure provisioning. AI agents that read a product specification, generate Terraform, validate it against cost and compliance policies, and apply it to a staging environment — without a human in the loop until the PR review.

Incident triage. Systems that page on an alert, correlate logs across 30 services, identify the probable root cause, and push a remediation recommendation to the on-call engineer — all in under 90 seconds.

Pipeline optimisation. Agents that analyse historical build times, identify flaky tests, reorder test suites for faster feedback, and submit the changes as a pull request with supporting metrics.

Cost optimisation. Automated rightsizing recommendations that don't just suggest but implement — resizing reserved instances, terminating idle resources, updating autoscaling policies — within defined guardrails.

Every one of these tasks used to require a skilled engineer. Most of them still do, today. By 2028, the majority won't.

---

What This Actually Means for Headcount

I'm not predicting mass unemployment. I'm predicting role compression and redistribution.

The DevOps engineer whose job is to write Terraform templates from requirements documents, respond to standard incident playbooks, and tune CI pipelines? That role shrinks dramatically. The AI does the execution. The human does the review, the policy setting, and the edge-case handling.

But the Platform Engineer who designs the guardrails within which AI agents operate? The Site Reliability Engineer who defines what "good" looks like and trains the incident response model? The Security Engineer who audits what the AI is provisioning and defines the blast radius? Those roles grow.

The demand for DevOps practitioners isn't disappearing. The definition of the job is changing.

---

The Uncomfortable Truth About "The Human in the Loop"

Here's where I'll push harder than most: the "human in the loop" framing is often a polite fiction.

When an AI agent raises a Terraform PR and a human rubber-stamps it in 30 seconds without reading it, the human isn't really in the loop. They're in the legal loop — providing cover for a decision the system already made.

This is going to be the dominant pattern for routine infrastructure operations within a few years. And that's fine, as long as we're honest about it and build the right governance structures around it — rather than pretending that approval-clicking is meaningful oversight.

Real human oversight means: reviewing the policies, setting the constraints, auditing the outputs, and catching the cases where the AI's optimisation objective misaligns with what you actually care about.

---

What the Engineers Who Survive Will Look Like

I've been hiring for 17 years. Here's what I'd look for in a DevOps engineer in 2028:

Systems thinking over tool mastery. The specific tools matter less. Understanding how distributed systems fail, how to design for reliability, how to reason about blast radius — those skills compound. Charity Majors, CTO of Honeycomb and one of the sharpest voices in observability, has argued for years that the senior engineers who deeply understand system behaviour are the ones automation will make more valuable, not less. She's right. The AI executes; the human has to know when the execution is wrong.

AI literacy as a first-class skill. The ability to evaluate what an AI agent did and why, to spot the confident-but-wrong hallucination in an infrastructure plan, to design prompts and guardrails that constrain agent behaviour effectively.

Policy and governance instincts. The boring compliance work that nobody wanted to do becomes critical when the AI is making hundreds of infrastructure decisions per day.

Business context. Knowing why the system exists and what the organisation actually needs, so you can define the right objectives for the systems doing the automation.

---

The Challenge

If your entire DevOps skill set is "I know Terraform and can debug Helm charts," you have about two years to retrain before that skillset is commoditised. Not worthless — but commoditised.

The engineers who'll thrive are the ones who stop being threatened by AI automation and start treating it as a productivity multiplier that frees them to work at a higher level.

The alternative is watching your job description slowly hollow out while pretending it isn't happening.

I've seen both responses. Only one of them ends well.

What's your read? Are you seeing AI automation hit your DevOps workflows? Book a call to discuss — I find these conversations genuinely useful.