A step-by-step path to become a DevOps, Platform or SRE engineer in 2026, with what is new this year and what to watch in 2027. Follow the main line top to bottom; click any topic to see why it matters and mark it as done.
Follow the main line from top to bottom. Each stage builds on the ones above it. You do not need to master a stage before moving on, but skipping the early ones always comes back to bite.
Solid green topics are the path. Learn those first. Outlined topics are alternatives: pick one per group. Dashed topics are good to know; come back to them when a job needs them.
Click any topic to read why it matters and where to learn it on this blog, then mark it as done. Your progress stays in this browser only.
Use the highlight buttons to see only the main path, what is new in 2026 or what to watch in 2027.
What DevOps means in 2026
DevOps is not a tool or a job title. It is a way of working where the people who build software and the people who run it share the same goals, the same pipeline and the same pager. In practice the role sits between development and operations: you automate builds and releases, manage infrastructure as code, keep systems observable and reliable, and make it easy for developers to ship safely.
In 2026 the job title varies (DevOps engineer, platform engineer, SRE, cloud engineer) but the skills overlap heavily. This roadmap covers that shared core and points out where each specialisation goes deeper.
What changed this year
AI is part of the toolbox. Assistants draft pipelines, Terraform and runbooks, and agents connect to Git, CI and clusters through the Model Context Protocol. The skill is no longer “can you write YAML” but “can you review and govern what was generated”.
Ingress NGINX was retired in March 2026. Gateway API is now the recommended way to route traffic into Kubernetes, so it moved onto the main path.
OpenTelemetry is the default. Instrument once and send traces, metrics and logs to whichever backend you choose.
No more long-lived keys in CI. OIDC federation between pipelines and clouds is the expected way to deploy.
Supply chain security went mainstream. SBOMs, signed images and provenance are requested by customers and regulators, not just security teams.
Platform engineering grew up. More companies run a platform team that offers golden paths and self-service infrastructure as an internal product.
Cost is an operational metric. FinOps practices sit next to latency and error rate on the dashboards.
What to expect in 2027
Predictions are guesses, but the direction is clear from what platform teams are building today:
Agents become platform users. AI agents will need their own identities, permissions, quotas and audit trails, exactly like human teams. Platform teams that already do RBAC and policy as code well will be ready.
Agentic operations, with people in the loop. Expect agents that triage alerts, propose fixes as pull requests and run approved runbooks. Analysts such as Gartner also expect many agentic projects to be cancelled for cost and risk reasons, so governance and measurable value matter more than demos.
Compliance as code. Audit evidence gathered automatically from pipelines, policies and Git history.
Post-quantum readiness and GreenOps. Inventories of cryptography in use, early hybrid TLS tests, and carbon reporting next to cost reporting.
A realistic 12-month plan
Months
Focus
Stages
1 to 3
Foundations: code, Linux, the terminal, Git and networking
1 to 6
4 to 6
Containers, one cloud and infrastructure as code
7 to 9
7 to 9
CI/CD, Kubernetes and GitOps
10 to 12
10 to 12
Secrets, observability, networking, reliability and security
13 to 17
Ongoing
Platform engineering, AI in operations and the 2027 topics
18 to 20
That pace assumes a few hours most days. If you already work as a developer or sysadmin, you can move through the first stages much faster.
Build these projects along the way
A Bash or Python script that checks disk, memory and a website’s health and posts the result to a chat webhook.
A small web app with a database in Docker Compose, then its images built and pushed by a CI pipeline.
The same app’s infrastructure on a cloud provider with Terraform, deployed from the pipeline with OIDC and no stored keys.
A Kubernetes cluster where Argo CD deploys the app from Git, with Gateway API routing and External Secrets.
OpenTelemetry, Prometheus and Grafana on top, with one SLO and an alert you would actually want to be woken up for.
Put each project in a public repository with a short README explaining your choices. That portfolio says more in an interview than any certificate.
FAQ
Do DevOps engineers need to code?
Yes, but not like product developers. You write scripts, automation, pipeline logic, infrastructure code and small tools, and you read application code well enough to debug it. Python plus Bash covers most of it; Go helps if you want to build cloud-native tooling.
What is the difference between DevOps, SRE and platform engineering?
DevOps is the culture and set of practices. SRE applies software engineering to reliability, with SLOs and error budgets at the centre. Platform engineering builds an internal product, the platform, that makes those practices self-service for developers. Most real jobs mix all three.
What does DevSecOps add, and what is shift-left?
DevSecOps makes security everyone’s job and builds it into the pipeline instead of a review at the end. Shift-left means moving checks such as secret detection, code scanning and dependency checks as early as possible, ideally into every merge request, where problems are cheapest to fix.
Which cloud should I learn first?
The one used by the companies you want to work for, and AWS if you have no preference because it appears in the most job ads. Learn one deeply; the concepts (identity, networking, compute, storage, managed services) transfer to the others in weeks.
How long does it take to become a DevOps engineer?
From zero, roughly a year of steady practice to be ready for a junior role, faster if you already work as a developer or sysadmin. The plan above is a realistic outline; real projects matter more than the calendar.
Will AI replace DevOps engineers?
It is replacing the boring parts: boilerplate YAML, first drafts of scripts, log summaries. It increases the demand for people who can design systems, judge whether generated changes are safe, set guardrails for agents and own the result when something breaks. Learn to use AI well and to govern it.
Are certifications worth it?
They help you get past filters and give structure to your study. The ones that carry weight are a cloud associate certificate (for example AWS Solutions Architect Associate), the Kubernetes CKA or CKAD, and the HashiCorp Terraform Associate. Pair them with projects you can show.
Which metrics show that DevOps is working?
The four DORA metrics: how often you deploy, how long a change takes to reach production, how often deployments fail and how quickly you recover. Add SLO compliance and cost per service, and you have a dashboard leadership understands.
Enough code to automate your own work, read your developers' code and write small tools.
Python
Learn it
The default glue language of operations: scripts, API clients, CLIs and most automation SDKs. Learn functions, modules, virtual environments and calling REST APIs.
Go
Alternative
The language of Kubernetes, Terraform, Prometheus and most cloud-native tools. Pick it second if you want to write operators, CLIs or contribute upstream.
JavaScript / TypeScript
Alternative
Useful when your team ships Node.js services or you write GitHub Actions, CDK stacks or Pulumi programs in TypeScript.
Rust
Good to know
Fast and memory-safe; more and more infrastructure tools are written in it. Good to read, rarely required for the job.
AI coding assistants
Learn itNew in 2026
Writing scripts, YAML and Terraform with an assistant is normal practice in 2026. Learn to give context, review every line it writes and never paste secrets into prompts.
02
Operating systems
Everything you deploy ends up on an operating system. Know it well enough to debug it at 3 a.m.
Ubuntu / Debian
Learn it
The most common Linux family in clouds and container base images. Learn apt, users and groups, permissions and the filesystem layout.
RHEL, Rocky and AlmaLinux
Alternative
The enterprise Linux family: dnf, SELinux and long support cycles. Common in banks, telcos and regulated industries.
systemd and services
Learn it
How services start, restart and log on modern Linux: units, timers, journalctl and resource limits.
Processes and performance
Learn it
Find what is slow or stuck: ps, top/htop, free, df, iostat, ss, lsof and strace. Know CPU, memory, disk and network pressure when you see it.
Windows Server
Good to know
Still runs a lot of enterprise workloads. Know services, event logs and remote management basics if your company uses it.
FreeBSD and other BSDs
Good to knowLegacy
Found in some appliances, firewalls and storage systems. Interesting, rarely a requirement.
03
Live in the terminal
Operations work happens in a shell. Get fast and comfortable there.
Bash scripting
Learn it
Variables, loops, conditionals, exit codes, set -euo pipefail and functions. Most pipeline steps are Bash in disguise.
grep, sed, awk, jq and yq
Learn it
Slice logs, rewrite config and query JSON or YAML without writing a program. jq alone saves hours every week.
Vim or Nano
Learn it
You will edit files on servers and in containers with no IDE. Know how to open, edit, save and quit at least one terminal editor.
SSH and keys
Learn it
Key pairs, agents, ~/.ssh/config, jump hosts and port forwarding. Prefer short-lived certificates or SSO-based access in production.
PowerShell
Alternative
The shell for Windows and Azure automation, and cross-platform since PowerShell 7.
tmux
Good to know
Keep sessions alive on remote machines and split one terminal into many panes.
04
Version control and code hosting
Everything is code now: apps, infrastructure, pipelines and policies. Git is how it all changes.
Git
Learn it
Commits, branches, merge versus rebase, resolving conflicts, tags and reading history. Practise undoing mistakes safely.
A list of every component you ship, in SPDX or CycloneDX, increasingly required by customers and regulators.
Signing and provenance (Sigstore, SLSA)
Learn itNew in 2026
Sign images with cosign and record how they were built, so clusters only run what your pipeline produced.
Runtime security with Falco
Good to know
Detect suspicious behaviour in running containers, such as a shell spawning in production.
Patching and CVE management
Learn it
A steady cadence for base images, dependencies and nodes, with bots such as Renovate opening the updates.
18
Platform engineering
Turn everything above into self-service paths that developers actually enjoy.
Internal developer platforms
Learn itNew in 2026
A product built by a platform team: templates, pipelines, environments and guardrails behind one interface.
Backstage and developer portals
AlternativeNew in 2026
A service catalog, docs and software templates in one portal. Backstage is the open-source standard.
Golden paths
Learn it
Opinionated, ready-made ways to create and ship a service that already include CI, observability and security.
Self-service infrastructure
Learn it
Developers request databases or queues through templates, Terraform modules or Crossplane, without filing a ticket.
Developer experience metrics
Good to know
Measure friction with surveys and flow metrics, not just uptime, and treat the platform as a product.
19
AI in DevOps
AI is part of daily operations in 2026. Use it well, run it safely and keep humans accountable.
AI assistants in daily work
Learn itNew in 2026
Drafting pipelines, explaining errors, summarising incidents and reviewing plans. Always verify the output before it ships.
MCP and tool-connected agents
Learn itNew in 2026
The Model Context Protocol lets AI agents call your tools (Git, CI, clusters, tickets) through a standard interface, now governed under the Linux Foundation.
AI governance and guardrails
Learn itNew in 2026
Decide what agents may read and change, use least-privilege credentials and keep an audit trail of every action.
Running models on Kubernetes
AlternativeNew in 2026
GPU node pools, model servers such as vLLM and KServe, and autoscaling for inference workloads.
LLM observability and evaluation
Good to knowNew in 2026
Trace prompts and responses, track cost and latency, and test model changes like any other release.
20
2027: what to watch
Where the field is heading next year. Learn the ideas now so you are ready when they reach your team.
AI SRE and agentic operations
Good to knowWatch in 2027
Agents that triage alerts, open fixes as pull requests and run approved runbooks, with people approving the risky steps.
Agents as platform users
Good to knowWatch in 2027
Platforms that give AI agents their own identities, quotas and policies, exactly like human teams.
Compliance as code
Good to knowWatch in 2027
Evidence for audits collected automatically from pipelines, policies and Git history instead of screenshots.
GreenOps
Good to knowWatch in 2027
Carbon-aware scheduling and energy reporting next to cost reporting.
Post-quantum cryptography
Good to knowWatch in 2027
Inventory where you use TLS and signatures, and test hybrid post-quantum key exchange as libraries and clouds adopt it.
WebAssembly workloads
Good to knowWatch in 2027
Small, fast-starting Wasm modules running next to containers for edge and plugin use cases.