How to become a DevOps engineer in 2026: the honest roadmap

Short answer: most people reach junior DevOps level in 8 to 14 months. Learn in this order: Linux, networking, Git, one cloud, CI/CD, Docker, Kubernetes, Terraform. Employers hire on demonstrated work, not certificates — three real projects beat five exams. No computer science degree is required.

Every DevOps roadmap on the internet is a list of twenty logos. That is not a roadmap, it is a shopping list, and it is why so many people spend a year learning tools and still cannot answer the only question an interviewer asks: tell me about something you deployed and what broke.

This article is the order I would learn things in today, with honest timings, and the reasons behind the order. It is written for the person who already works in tech — support, QA, development, sysadmin — and wants to move.

What does a DevOps engineer actually do?

Strip away the job titles and it is three things: make software reach production automatically, keep it running, and make both cheap enough that nobody panics at the invoice.

In practice, a normal week involves writing pipeline code so a merge becomes a deployment, describing infrastructure in files rather than clicking in a console, investigating why something is slow at 3pm, and cutting a cloud bill somebody has stopped reading. Very little of it is "collaboration culture", whatever the conference talks say. It is engineering, and its raw material is other people's software.

The role also splits as you go up. Junior work is pipelines and tickets. Mid-level is Kubernetes, infrastructure as code, cloud architecture. Senior work is platform design — building the paved road other developers walk on — and that is where the title often changes to platform engineer or SRE without the work changing much.

How long does it take to become a DevOps engineer?

It depends almost entirely on what you already have, so here are the honest brackets.

Where you start	Realistic time to junior-ready	What takes the longest
Developer, comfortable with Git and the terminal	8 to 12 months	Infrastructure, networking, production instincts
Sysadmin or IT support with Linux experience	6 to 10 months	Coding, CI/CD, infrastructure as code
QA or technical support, some scripting	12 to 16 months	Everything at once — go slower on purpose
Career changer, no technical background	18 to 24 months	Build the developer foundation first

Two things move these numbers more than talent. The first is hours per week: ten hours changes the outcome, three hours does not. The second is whether you finish things. Six half-built projects teach less than one project deployed, broken, and repaired.

The order to learn things, and why the order matters

The sequence below is not arbitrary. Each step is the thing you need in order to understand the next one, and skipping ahead is the reason people bounce off Kubernetes and conclude they are not smart enough.

1. Linux and the terminal — 3 to 4 weeks

Everything you will later automate runs on Linux. You need file permissions, processes, systemd, logs, package management, and enough Bash to write a script you are not ashamed of. The test: you can diagnose a service that will not start using only journalctl, ps and ss.

2. Networking — 1 to 2 weeks

DNS, TCP/IP, ports, HTTP, TLS, SSH, what a load balancer does. Almost every "Kubernetes is broken" incident is a networking problem wearing a costume. Skipping this is the single most expensive shortcut in the list.

3. Git, properly — 1 week

Not just add, commit, push. Branching strategies, rebase versus merge, resolving conflicts without panic, pull request review, and what .gitignore saves you from. Everything downstream — pipelines, infrastructure as code, GitOps — is Git with extra steps.

Our Git and GitHub course is free with an account, and it is the right week-one investment because the rest of the roadmap assumes it.

4. One cloud, deeply — 6 to 8 weeks

Pick AWS unless your local market says otherwise. Learn compute, networking (VPC, subnets, security groups), storage, IAM, and cost. Depth in one cloud beats surface knowledge of three, and IAM is where beginners lose the most days, so do not skim it.

5. Containers — 2 weeks

Docker images, layers, multi-stage builds, volumes, networks, Compose. Build an image that starts in under a second and does not run as root. That last detail comes up in interviews more than you would expect.

6. CI/CD — 2 weeks

GitHub Actions or Jenkins. A pipeline that tests, builds an image, pushes it to a registry, and deploys. Then make it fail on purpose and fix it. Pipelines are the part of the job you will touch every single day, which is why the Jenkins and CI/CD course sits in the middle of the track rather than at the end.

7. Kubernetes — 6 to 8 weeks

Pods, deployments, services, ingress, config and secrets, RBAC, resource limits, Helm. This is the longest step and the one that pays best. Run it locally first (kind or minikube) before touching a managed cluster, because a local cluster you can destroy teaches faster than a cloud bill you are afraid of.

Kubernetes is free with an account here, lessons and labs included, for exactly that reason: it is the gate everyone has to walk through.

8. Infrastructure as code — 3 to 4 weeks

Terraform. Modules, state, plan and apply, remote backends, and the discipline of never clicking in a console again. If you want to practise without spending anything, Terraform with LocalStack runs the whole loop against a local AWS emulator.

9. Observability, and only then everything else

Metrics, logs, traces, alerts that mean something. Prometheus and Grafana, or whatever your cloud provides. After this, add security, cost management and platform work as your job demands them — not before.

Do I need a degree or certifications?

No degree. The market for infrastructure roles is one of the most credential-agnostic in software: most working DevOps engineers arrived from development or system administration, not from a curriculum.

Certifications are a different question. They will not get you hired on their own, but they do three useful things: they get you past keyword filters, they force a syllabus on people who drift, and they prove you finished something. Three are worth the money in 2026:

• CKA (Certified Kubernetes Administrator) — the one that carries weight, because it is a hands-on exam in a real terminal. Budget four to six weeks of dedicated practice beyond general familiarity.
• HashiCorp Terraform Associate — cheap, quick, and it maps exactly to the work.
• AWS Solutions Architect Associate — the broadest cloud credential, and the one that appears most often in job postings. If you are unsure which AWS exam to start with, we wrote a full comparison of the AWS certification path.

Two or three well-chosen certificates plus a portfolio beats a wall of badges. Nobody has ever been hired for their eleventh certificate.

What should be in the portfolio?

Three projects, each finished and each documented in a README that explains the decisions, not just the commands:

1. A pipeline. A small application, containerised, tested and deployed automatically on every merge, with a rollback that works.
2. Infrastructure as code. The same application's infrastructure in Terraform — network, compute, database, secrets — that can be destroyed and rebuilt from zero.
3. A cluster you operate. The application on Kubernetes with health checks, resource limits, an ingress, monitoring, and a written post-mortem of one failure you caused deliberately.

That third document is the one that gets you hired. Anyone can follow a tutorial. Writing "I set memory limits too low, pods were killed, here is how I found it in the events and what I changed" is what a hiring manager is actually looking for.

The four mistakes that stall people

Learning twelve tools at once. One thing, finished, then the next. The list above is sequential for a reason.

Reading instead of breaking. You do not learn Kubernetes by reading about Kubernetes. You learn it by deleting something important at 11pm and getting it back.

Skipping Linux and networking because they feel less exciting than the cloud console. Every hour you skip there costs three later.

Treating AI as an answer machine. Use Copilot, Claude or Cursor from day one — engineers who refuse them are slower now, and that is simply true. But read every line before you run it, and write some things without help. The failure mode of 2026 is the candidate who ships fast and cannot debug anything.

What does a DevOps engineer earn?

In North America, entry-level roles land roughly between 85,000 and 105,000 US dollars, mid-level between 118,000 and 145,000, and senior between 165,000 and 190,000, with wide variation by city and industry. In Europe the numbers are lower in absolute terms and much more variable between countries. Remote roles pay closer to the company's market than yours, which is the strongest financial argument for building a public portfolio.

A 12-month plan you can actually follow

• Months 1 to 2: Linux, networking, Git. Automate something boring in your current job.
• Months 3 to 4: AWS fundamentals, IAM, VPC, cost. Build and destroy real infrastructure by hand.
• Months 5 to 6: Docker and CI/CD. Project one goes live.
• Months 7 to 9: Kubernetes locally, then managed. Project three. Take the CKA if you want a deadline.
• Months 10 to 12: Terraform, observability, and the job hunt: resume, portfolio README files, interview practice.

Interviews are their own skill, and it is a learnable one. Our tech job search course covers the part most engineers get wrong — not the technical questions, but everything around them.

Frequently asked questions

Can I become a DevOps engineer with no IT experience?

Yes, but plan for 18 to 24 months rather than 12, and build a developer or sysadmin foundation first. DevOps automates software delivery, so it is hard to automate a process you have never taken part in. The fastest route from zero is usually a support or junior development role while you learn the infrastructure side in the evenings.

How long does it take to learn DevOps?

Eight to fourteen months for someone already working in tech, studying ten or more hours a week. Developers who know Git and Linux can reach junior readiness in six to eight months. Career changers with no technical background should budget two years.

Do I need to know how to code to work in DevOps?

Yes, though less than a software engineer. You need Bash for glue, Python or Go for tooling, and the ability to read the languages your teams use well enough to debug a failing build. You will write more YAML and HCL than application code. Python is free with an account here.

Which DevOps certification is worth it in 2026?

The CKA carries the most weight because it is a hands-on exam. The HashiCorp Terraform Associate is cheap and maps directly to daily work, and the AWS Solutions Architect Associate is the most requested cloud credential. Two or three, alongside real projects, is the sweet spot.

Is DevOps still a good career in 2026?

Yes, though the title is drifting toward platform engineering and site reliability engineering. The underlying demand — automating delivery, running distributed systems, controlling cloud cost — has grown rather than shrunk, and AI tooling has increased the amount of software that needs deploying and monitoring.

Where should I start today, with no money?

Git, Python and Kubernetes are free with an account on this site, lessons, labs and quizzes included, and they cover three of the nine steps above. Add a cloud free tier and you can reach the middle of the roadmap without paying anything. See the free courses.

Where to go next

If you want the sequence as a visual path rather than a list, the learning roadmap lays out the same progression course by course. And if you would rather start tonight, start with Git — it is free, it takes a week, and everything else in this article assumes it.