detail A · it is a product now
AWS DevOps Agent became generally available on 31 March 2026, priced at $0.0083 per agent-second. Teams now decide whether to buy a product like it or build their own, and this course teaches you to judge both.
a hands-on course, drawn from a real incident
Hands-On AIOps on AWS: Build AI Agents for DevOps
You build an agent that reads the alert, finds the cause, proposes one exact fix and waits for two people to approve it. Then you measure how often it is right. The drawing on this page is a real run from 6 November 2026.
One payment, lifetime access. It rises to ₹7,777 or $177 once every part is released, and everything published later is included.
- The opening pages, Part 1 and most of Part 2 are available now
- New parts every week
- Text lessons with real screenshots and code downloads
- Video being added
Written by Roni Das. Every number on this page comes from a recorded run, and the free pages show you his teaching before you pay.
- alerts receivedThree alerts fire on the demo shop and are grouped into one serious incident.
- investigatingThe agent makes 6 read-only checks in 11 seconds and changes nothing.
- cause foundThe image tag nginx:1.29.99-alpine does not exist, so the web pod cannot start.
- waiting for approvalIt proposes one exact fix, then stops and waits for people to approve it.
- fixingTwo people approve, and the Executor applies only that one change.
- resolvedThe shop is healthy again. The whole incident used 7 model calls and cost $0.00055.
the drawing above, played back
Replay the real incident step by step, and approve the fix yourself
Every line below is from the terminal of the run on 6 November 2026, and every number from its record of model calls. Press play to follow the agent from the alerts to the cause in 11 seconds. At the approval it stops and waits for two people, and here those people are you.
The replay loads as you reach it.
drawing 01 · why this agent
Why an AI agent for on-call work is the right first agent project to build
Hands-On AIOps on AWS is a self-paced, project-based course by Roni Das on systemdesign.academy: you build AI agents for DevOps incident response. You build an AI agent that reads an alert, investigates a Kubernetes system with read-only tools, finds the cause, proposes one exact fix and waits for a person to approve it, and then you measure how often it is right. It is for developers who already know Python and the command line. You build it on your laptop first, using OpenAI models, and then run it in production on AWS: EKS with one IAM role per agent, real AWS incidents and an audit trail.
Incident work teaches agents properly: the agent has real data to read, every step can be checked, and a wrong action has a real cost. So the design keeps the model away from any change. Follow the numbers in the drawing.
detail B · it is not solved
On ITBench-AA, published on 27 May 2026, no AI model scored above 50% precision at finding all the root causes across 59 injected Kubernetes faults. That is why this course measures the agent instead of trusting it.
detail C · the real numbers
The run in the drawing above found the cause in 11 seconds with 6 read-only tool calls, and the whole incident cost $0.00055 in model fees. You see these numbers for every run you make.
- drawing
- Who may change what during an incident
- checked by
- the evaluation harness, not a promise
- source
- the recorded run, 6 Nov 2026
what you build, part by part
What you build in this course: the whole AI on-call system, on your laptop and then on AWS
Tap or click a number, or a part in the list, to see where it sits. Parts 1 to 5 run on your own laptop; only Parts 6 and 7 go to AWS.
What you leave the course with
- A working agent system in your own repository: the agent, the Approval Gate, the Executor and the Ops Console, running on your laptop and on AWS.
- Measured results: an evaluation report on the 30 incidents with root cause, wrong actions and cost, including where the agent still fails.
- An answer for interviews and for work: how to build an agent that is safe to put near production, and when to buy a product such as AWS DevOps Agent instead.
your path through the course
From your first agent to an AI on-call team on AWS that you have proved
Each step below is a real part of the course, with its page count. Parts 1 to 5 run on your own computer, apart from one Part 1 page that reads a real CloudWatch alarm; Part 6 takes the team to AWS, and the last part proves it works.
- Part 1Agents from zero13 pages
- LangGraphPart 2LangGraph from zero10 pages
- Part 3One DevOps agent, done safely and measured8 pages
- MCPPart 4Tools the standard way, MCP6 pages
- teamPart 5From one agent to a team9 pages
- Part 6Production on AWS7 pages
- You are an AIOps engineerProve it: 5 pages, ending in a final evaluation you can show.
The 30 real incidents you break on purpose, so the agent learns to find their cause
Every incident is a fault you inject yourself, with a known cause and an automatic check. Tap a note to turn it over. Number 1 is the one in the drawing at the top of this page.
- And the trap casesA change during a freeze, or a “fix” that would delete data. The right answer is to refuse and page a person, and the agent is scored on that too.
- And the attacksInstructions hidden in a log line, a GitHub issue, a pod annotation, a tool description and a planted memory. You attack your own agent and measure the block rate.
The contents of AIOps on AWS, part by part, and which pages you can read free today
The course is laid out like a book: 70 pages in 7 parts, three opening pages and an optional appendix. Available now: the opening pages and Part 1 and Part 2. New parts are added every week, in book order.
Start here: the opening pages
The whole course in one real incident, what you should already know, a one-screen setup checklist, and your model key.
3 of 3 pages available now
- Introduction: what this course buildsfree, read now
- Before you startfree, read now
- Getting a model API key and keeping it safeavailable
Part 1: Agents from zero
What a language model really does, how an agent works, and how to build one safely, first by hand and then with LangChain.
13 of 13 pages available now
- What a language model really doesavailable
- Talking to a model: messages, roles and temperatureavailable
- Answers you can trust in codeavailable
- What an AI agent isfree, read now
- One agent loop, step by stepavailable
- Tools for an AI agentavailable
- Designing tools and contextavailable
- Write the agent loop yourselfavailable
- The same agent with LangChain's create_agentavailable
- Give the agent an AWS toolavailable
- Agent middleware: limits, retries and approvalavailable
- Decision models for fixed choicesavailable
- A cost log for every model callavailable
Part 2: LangGraph from zero
Building agents as graphs: shared state, loops, pausing for a human, surviving a crash, memory and tracing.
10 of 10 pages available now
- Why build an agent as a graphavailable
- LangGraph state and reducersavailable
- Routing, loops and the recursion limitavailable
- Hand-offs, parallel work and streamingavailable
- Durable agents that survive a crashavailable
- Asking a human with interruptavailable
- Timeouts, error handlers and cachingavailable
- Agent memoryavailable
- Deep Agents, a ready-made harnessavailable
- Tracing an agent with LangSmithavailable
Part 3: One DevOps agent, done safely and measured
The first real agent on the real shop, with an approval gate it cannot get around, an Ops Console, and honest numbers.
8 pages, coming
- The demo shop we break all coursecoming
- The Kubernetes agent on the real shopcoming
- The Approval Gate and the Executorcoming
- The Ops Consolecoming
- The first attack: an instruction hidden in a log linecoming
- What it means for an incident agent to workcoming
- The evaluation harnesscoming
- Fair numbers for an agentcoming
Part 4: Tools the standard way, MCP
The Model Context Protocol, real MCP servers, a gateway that controls which agent may call which tool, and MCP security.
6 pages, coming
- What MCP iscoming
- Our tools as an MCP servercoming
- Real MCP servers for Kubernetes, GitHub and Grafanacoming
- Your own MCP server for security scanningcoming
- The MCP gatewaycoming
- MCP securitycoming
Part 5: From one agent to a team
Team patterns, an Incident Commander, one specialist per area, a Verifier and a Scribe, and what a team really costs.
9 pages, coming
- Multi-agent team patternscoming
- What a team of agents costscoming
- Designing the context firstcoming
- The Incident Commander agentcoming
- The specialist agentscoming
- The Verifier and the Scribecoming
- When an agent team goes wrongcoming
- Alert intake and paging a humancoming
- Long-term memory for the teamcoming
Part 6: Production on AWS
AWS with cost safety first, Terraform, EKS with one IAM role per specialist, the AWS incidents, and an audit trail.
7 pages, coming
- AWS from zero, cost safety firstcoming
- Terraform, practised on the laptop firstcoming
- Amazon EKS with one IAM role per agentcoming
- The Cloud agent and the AWS incidentscoming
- Kyverno, Secrets Manager and an audit trailcoming
- OpenTelemetry export inside your own networkcoming
- Approving agent changes from Slackcoming
Part 7: Prove it
An operations report, a red team against your own agents, the final evaluation, and a comparison with AWS DevOps Agent.
5 pages, coming
- The operations reportcoming
- Red-teaming your own agentscoming
- The final evaluationcoming
- Buy or buildcoming
- Capstone: a new fault and a new specialistcoming
Appendix: DevOps basics, if you need them
Optional. Working DevOps engineers can skip it; everyone else can read a page whenever a part of the course needs it.
9 pages, coming
- Containers from zerocoming
- Kubernetes podscoming
- Kubernetes deploymentscoming
- Kubernetes servicescoming
- Kubernetes health probescoming
- Kubernetes storagecoming
- Kubernetes autoscalingcoming
- Observability from zerocoming
- CI/CD and GitOpscoming
Who this AIOps course is for, and what you should know before you start
Before you start, you should know
- Python: you can write small programs
- The command line: you are at home in a terminal
- Git basics: clone, commit, push
You do not need to know
- AI or agents: Part 1 starts from what a language model really does
- Kubernetes or DevOps: an optional appendix covers them
The free page “Before you start” has a three-question self-check and a one-screen setup list.
Roni Das writes the System Design Masterclass and the AI Engineering course on systemdesign.academy, 989 lessons in all, and two full-length handbooks on freeCodeCamp. Every run, screenshot and number in this course comes from a real run, and each page says when it was recorded.
What readers say about Roni Das's other course (System Design Masterclass: 5.0 out of 5 from 5 published reviews). AIOps on AWS is new and has no reviews yet. Read every published review.
- ★★★★★
Ive started this course by accident - LLM advised me this site. When I passed few free lessons, I had no doubt - this course should be bought. And I did it with special pleasent price and it was best decision of the month. Content is great, interesting to read and with a lot of practial cases (what makes it different from other souces and books) I hope to teach all the lessons, that will be my supergoal. Glad that I found this resource and speacial Thank You to Roni Das - Great Job!
Oleksander, Backend Engineer · System Design Masterclass - ★★★★★
The material has been designed with a lot of effort and dedication. It has covered vast range of topic with sufficient content and diagrams wherever required. It complements well with other system design materials.
Som · System Design Masterclass - ★★★★★
Due to time constraints I am not regularly reading it but whenever I get a chance I read the concepts which have been explained very well. For few topics I felt some more details would be beneficial but I am sure Roni will take care of that because I noticed that the content is getting refreshed and new content is getting added all for free with only one time small fee for lifetime access.
Ajay Arora · System Design Masterclass
How the course is delivered: written pages, code you can run, and video as it is added
A written page for every lesson
Every command, real screenshots from real runs on a real machine, diagrams, and notes for Windows.
Code you can download
A zip with every page that has code. What you see run is exactly what you download and run yourself.
Video, being added
Videos for the opening pages and Part 1 are recorded and are being added to the pages. Later parts come as text first, and their videos follow part by part.
What it costs to run
The recorded run cost $0.00055 in model fees. Parts 1 to 5 need no servers. On AWS, from Part 6, a 4-hour lab is about $1.50 to $3 if you tear it down after.
The early-access price of AIOps on AWS, and why it is lower now
The course is being released part by part, so it sells at an early-access price today. Once every part is released, the price rises to ₹7,777 or $177. If you buy now, you pay once and get every part and every video as it comes out, at no extra cost.
Not sure yet? Read the free pages first: the introduction, “Before you start” and “What an AI agent is”. They are there so you can judge the teaching before you pay.
Read the free pagessystemdesign.academy
receipt · AIOps on AWS
- 70 pages in 7 parts
- included
- Parts published later
- included
- Code zips and videos
- included
- Full price, later
- ₹7,777 or $177
- You save today
- ₹4,444 or $77
Total, one payment₹3,333 or $100
lifetime access
Buy early accessIn India you pay in rupees through Razorpay, by UPI, card or netbanking. Outside India you pay in US dollars by card through Razorpay, or with PayPal.
$ questions people ask before buying the AIOps on AWS course
? What do I need to know before I start this AIOps course?
You should be comfortable with Python and the command line, and know the basics of Git; the course does not teach those. You do not need to know AI or DevOps. Part 1 starts from what a language model really does and builds agents from zero, and an optional appendix teaches containers and Kubernetes for anyone who has not used them. The free page "Before you start" has a short self-check and a one-screen setup list.
? Is AIOps on AWS a video course or a text course?
Both. The course is laid out like a book of parts and pages, and every page is a written lesson with real screenshots from real runs and diagrams. The videos for the opening pages and Part 1 are recorded and are being added to the pages; later parts are published as text first, and their videos follow part by part. Each page with code comes with a zip of that code.
? Are all the parts available on the day I buy?
Not yet. The course has 70 pages. Today 26 of them are published, and 3 of those are free to read before you buy; new parts are added each week, in book order. That is why it sells at an early-access price, ₹3,333 in India or $100 elsewhere, which rises to ₹7,777 or $177 once every part is released. Everything published later is included in the one payment, so you get each new part and video at no extra cost.
? How much will the model API cost me while I learn?
Very little. Our recorded incident run on 6 November 2026 made 7 model calls and cost $0.00055. Our full test of the first version on 5 November 2026, 220 incident runs on two models, one of them a larger model used only for comparison, cost $2.55 in model fees in total. The page "A cost log for every model call" builds a cost log, so you always see what each run cost.
? Do I have to pay for AWS to do this course?
Only from Part 6, the production part. Parts 1 to 5 need no servers: they run on your own laptop, and you pay only the small model fees. On AWS, a 4-hour lab session costs about $1.50 to $3 if you tear it down afterwards, and about $8 to $9 a day if you leave it running. The first page of Part 6, "AWS from zero, cost safety first", sets up a budget alert and a one-command teardown before anything else.
? Does the agent change my systems on its own?
No. The agent only reads, with read-only tools. When it wants to change something, it proposes one exact change and stops. A person approves it, and then our own plain code, the Executor, applies exactly that change and nothing else. You build this approval flow yourself in Part 3.
? Who are the two people who approve a change when I am learning alone?
In the lab you type both names yourself, as in the recorded run, where the approvers were "student" and "reviewer". The Approval Gate checks that two different names approve and that neither of them is the agent, which is the same rule a team uses at work. In Part 6 you can move the approvals into Slack, where the two names are real people on your team.
? What computer do I need for the AIOps course?
A Windows or Mac laptop. Windows works through WSL2, and the free page "Before you start" lists the setup. The full system in the later parts (the demo shop, the cluster and the agent running together) needs about 16 GB of memory; the early parts are lighter. You also need an OpenAI API key, and the page "Getting a model API key and keeping it safe" shows how to create one with a spending limit.
? How is this different from using AWS DevOps Agent?
AWS DevOps Agent is a managed product from AWS, generally available since 31 March 2026 and priced at $0.0083 per agent-second. In this course you build your own agent, so you understand every step it takes and can measure it. The page "Buy or build" puts your agent next to AWS DevOps Agent and the agent SDKs, so you can judge when to buy and when to build.
? Is this course included in the System Design or AI Engineering course?
No. AIOps on AWS is its own course with its own price. Buying it gives lifetime access to all of its pages, videos and code, including every part published later.
? How do I pay for AIOps on AWS?
In India you pay in rupees through Razorpay, by UPI, card or netbanking. Everywhere else you pay in US dollars by card through Razorpay, or with PayPal. The pages unlock on your account as soon as the payment is confirmed.