Example · DevOps / Platform Engineer
63% of your week as a DevOps engineer is still yours.
AI now writes infrastructure code and pipelines. Answering for each production change stays with you.
Make my planA typical week at the middle pace. Your plan uses your own tasks.
Is DevOps a dying career?
In a typical week at the middle pace, 35% of this work needs your judgment and 28% is work you do using AI. AI could do about 37%. A model already does 37% of your week well, and that is the part you were hired for. That includes the Terraform module from a ticket, the pipeline workflow, the runbook and the access request queue.
Read more
If you have let an AI agent draft a module, you have seen it. The tags and parameters come out cleaner than yours. You may also have seen it propose a security group open to the world.
That second part is now the core of the job. Someone has to read the plan before it applies and say go on a production change with their name on the approval record. Someone has to decide at 3 a.m. whether to roll back, and tell finance why the bill went up and what to do about it. The work the model does well is cheap to produce and expensive to get wrong, and you are the person who knows where the bodies are buried in those accounts. The night page and the access queue are often the things people in this role dread most. They are also the most protected things you do. This report helps you take charge of those two areas.
The week, task by task
Who does each hour of a typical 40-hour week.
See every task
Needs your judgment.
- On-call and incident response: command, rollback, failover, comms12%
- Production change approval and deployment windows7%
- Cloud cost review: the monthly bill, rightsizing, commitment decisions6%
- Platform and architecture decisions: environment strategy, new-service onboarding, migrations6%
- + 1 smaller task4%
AI does the slow parts. You still own it.
- Alert triage, first-pass log reading, noise tuning8%
- Kubernetes upgrades, node pools, chart bumps8%
- Developer support: "my deploy failed," pairing on pipelines7%
- Audit evidence: access reviews, screenshots, assessor requests5%
AI could do most of it.
- Writing infrastructure-as-code modules and resources from a ticket12%
- CI/CD pipeline work: workflows, build fixes, caching, matrix jobs10%
- Routine access and IAM requests: bucket policies, role grants, key rotation, offboarding9%
- Runbooks, docs, change records, postmortem first drafts6%
Why some tasks still need a person
Deciding at 3 a.m. whether to roll back, fail over or wait
Someone must be reachable, authorized and willing to be wrong in public. An agent can propose a fix, but it cannot be paged, and no company yet lets one hold production write credentials unattended. The on-call rotation is the last place where a named human is required.
Approving a production change and being the name on the record
SOC 2 (a common security audit) has a change-management criterion (CC8.1). It requires that changes be authorized, tested and approved before they reach production, with evidence. Your auditors need a named approver. Today that is often your lead. The plan is about making it you.
Catching the open-to-the-world rule in a generated plan
A wrong infrastructure plan looks like a right one until the breach. The reviewer who reads the diff and knows what an open-to-the-world network range costs is the constraint on every infrastructure agent.
Knowing which cluster a service actually runs in, and why the dependency map is wrong
Past migrations, exceptions and "we'll fix it later" live in your head and nowhere else. A model can read the code, but it cannot know about the service that was moved by hand and never codified.
Explaining the bill and deciding what to do about it
The monthly cost number is read by finance. Rightsizing a cluster means knowing which workloads are latency-sensitive and which teams will object. A model can rank candidates, but a person has to choose and answer for it.
Good to know
What is hard about this field, and what you can do about it.
The hard truth
AI was supposed to cut your toil, and toil rose.
In Catchpoint's 2025 survey, toil rose for the first time in five years. More tools and faster changes can mean more to run and clean up.
Sources (2)
- Median time on operations work, often synonymous with toil, rose from 25% to 30%, the first rise in five years. Catchpoint, SRE Report 2025, 2025
- AI adoption still has a negative relationship with software delivery stability. Google Cloud, 2025 DORA Report, 2025-09
What's next for a DevOps or platform engineer?
Plan A is the best move inside the work you do now.
Grow in the job you have
The main focus. Most weekly sessions build toward it.
- Each week
- 4 sessions
- Ready in
- About 5 weeks for your first signed change note
Why it fits
Why you. If you already say "roll back" at 3 a.m. and already catch the open-to-the-world rule, a record of production changes and incidents handled is standing nobody else at your level has. The gap between what you do and what your title says is smaller here than for your other moves.
The gap. Most teams have no written severity ladder. The approval record names your lead, and incident command falls to whoever is on call. There is usually no plan-review checklist and no automated policy check.
The trade-off. You become the person blamed, in writing, when a production change goes wrong. The dread is the same, but you have more authority on paper. You will also merge fewer modules for a quarter, and someone will count.
2 other moves that fit
- Own the bill: the platform cost seatReady in about 7 weeks for your first cost memo
- The portable hedge: build a paper trail you can take with youReady in about 10 weeks to know your market rate
What's a plan B for DevOps engineers?
Plan B is meant to be different from this job. It was picked because skills from this role carry over.
Get ready for a new kind of job
A second option. A newer job AI is creating that these skills already fit.
Answering for production changes and running incidents is the heart of this role, so shipping one agent workflow gets you most of the way.
- Each week
- 1 session
- Outlook
- Hiring now
- Pay
- $200K to $485K a yearThree staff and principal engineer postings, Aug to Sept 2026; the top end is AI lab pay.
2 other options
- AI cost and vendor analystSince you already own cloud cost and tagging, adding model and GPU spend to the bill you defend is a natural extension.
- Developer AI enablement leadPlatform teams already shape how other engineers ship, and rolling out coding agents across your org is the proof this role asks for.
In your own plan you can switch to any of the 47 roles.
The 13 weeks
Small steps that add up. One plan for both: most sessions build Plan A, one a week builds Plan B.
By week 13 you will have
AFor Plan A
- A plan-review checklist and severity ladder in use
- A one-page memo with your ask for a Senior title
- A published write-up on agent reliability
BAnd for Plan B
- A regression incident report
- A published agent reliability write-up for your portfolio
- 1Measure your week
- 2Check the rules
- 3Start an incident ledger
- 4Write a plan-review checklist
You finish withA log of every on-call decision this week
- 5Write your first change-risk note
- 6Turn a rule into code
- 7Write a cost memo
- 8Write an access standard
You finish withA second log of on-call decisions
- 9Write a second note and ask
- 10Check the job market
- 11Gather your evidence
- 12Make your ask
- 13Choose your next step
You finish withOne application with your incident story attached
Three courses from the plan
- AWS Certified DevOps Engineer - Professional (DOP-C02)AWS Training and Certification · $300
Not during the plan. It certifies the exposed rows. Sit it only if the postings you want ask for it three times.
- AWS Certified Solutions Architect - Associate (SAA-C03)AWS Training and Certification · $150
Only if you have no AWS cert at all and an HR screen needs one. It tests vocabulary more than judgment.
- HashiCorp Certified: Terraform Associate (004)HashiCorp · $70.50 plus tax; no free retake
It is cheap and harmless, though it won't protect your job. If you sit any exam, sit this one.
Checked 2026-10-11.
The fuller picture
Risks beyond AI: moderateCost-center risk
Platform teams are an engineering cost center and were trimmed with everyone else in the 2023 tech layoffs. Pulling the other way, cloud spending keeps growing and the bill needs an owner. This risk does not outrank the AI risk for you, but it decides when a thin platform team gets thinner.
Where this could be wrong
- On-call may be more exposed than we scored. Our score assumes no vendor lets an agent roll back production unattended. If your company adopts automated rollback when a service misses its reliability target (its SLO) and trusts it, on-call becomes more exposed and the need for a person on the pager shrinks. The change and incident owner move then rests on change approval alone.
- We assumed the approval record names your lead. If a release manager or squad leads already sign and you only apply the change, the change and incident owner move is "join the approval," not "own it," and owning approvals protects you less.
- The access queue may already be on its way out. If your company adopts just-in-time access with automated approval, the access queue shrinks on its own and the access standard in Week 8 is moot. Spend that week on the cost memo instead.
- A staff engineer could take the change and incident owner move quickly. A senior engineer who already holds the agent's guidelines has a strong claim to it. If they want it, focus on the cost seat and the portable hedge.
Also matched: devops engineer, platform engineer, cloud engineer. Research checked 2026-10-11. Page updated .