How to Pass AI-300: A 6-Week Study Plan

To pass AI-300, you need to run two kinds of AI in production: classic machine learning on Azure Machine Learning, and generative AI apps and agents on Microsoft Foundry, with DevOps tools like GitHub Actions and Bicep woven through both. This guide gives you a six-week plan mapped to Microsoft’s official domain weights, a hands-on lab checklist, and the exam-day details that trip people up.

AI-300 leads to the Microsoft Certified: Machine Learning Operations Engineer Associate credential and replaced DP-100 on May 12, 2026. Still deciding whether it’s worth your time? Read our honest look at AI-300 first.

AI-300 at a glance

What the domain weights tell you

The study guide’s five domains and their weights tell you where your hours should go:

Two things stand out. The model lifecycle domain is the biggest block, so it gets two weeks. And the three generative AI domains add up to 40–55% of the exam, so DP-100 study material alone won’t get you there. Our DP-100 retirement explainer covers what changed between the two exams.

Are you ready to start?

Microsoft expects a data science background with Python, plus entry-level DevOps skills: GitHub Actions, command-line tools, and infrastructure as code with Bicep and the Azure CLI. Quick self-check: if you can write a Python training script, commit it to GitHub, and read a YAML workflow file, the six-week plan fits. If Git or YAML is new to you, add a week zero for GitHub basics first.

The free official resources

• The study guide. The AI-300 study guide lists every skill bullet under each domain. Treat it as your checklist.
• Two free learning paths. The self-paced version of course AI-300T00 is two Microsoft Learn paths, Operationalize machine learning models (MLOps) and Operationalize generative AI applications (GenAIOps), at roughly six hours each. It’s also taught as a four-day instructor-led course.
• The practice assessment. A free official AI-300 practice assessment lives on AI Skills Navigator; you must be signed in to launch it. It shows the style and difficulty of the exam, not the real questions.

One cost to plan for: the exercises run in real Azure. The AI evaluation lab, for example, requires an Azure subscription where you have administrative access, plus a GitHub account. Set a budget alert and delete each lab’s resource group when you finish, because running compute and endpoints keep billing.

The 6-week AI-300 study plan

This plan assumes 6 to 8 hours a week: a few evenings of reading plus one longer weekend lab. Each week lists the study guide skills and the matching Microsoft Learn modules.

Week 1: Workspace setup and infrastructure as code (15–20%)

• Create and manage a workspace, datastores, and compute targets, and set up identity and access.
• Create data assets, environments, and components; share assets across workspaces with registries.
• Deploy a workspace with Bicep and the Azure CLI, automate provisioning with GitHub Actions, and restrict network access.
• Modules: Get started with machine learning in Azure and Experiment with Azure Machine Learning.

Week 2: Training, tracking, and pipelines (lifecycle domain, part 1)

• Track experiments with MLflow and compare model performance across jobs.
• Use automated ML, run training scripts, and automate hyperparameter tuning.
• Build training pipelines, and know when large or deep learning models need distributed training.
• Modules: Run training scripts and track models with MLflow in Azure Machine Learning, Perform hyperparameter tuning with Azure Machine Learning, and Run pipelines in Azure Machine Learning.

Week 3: Registration, deployment, and monitoring (lifecycle domain, part 2)

• Register an MLflow model, evaluate it against responsible AI principles, and archive old versions.
• Deploy to real-time and batch endpoints, then test and troubleshoot them.
• Practice progressive rollout and safe rollback.
• Detect data drift, monitor production metrics, and set retraining or alert triggers.
• Modules: Automate model training with GitHub Actions and Deploy and monitor a model in Azure Machine Learning.

Week 4: GenAIOps infrastructure on Microsoft Foundry (20–25%)

• Create Foundry resources and projects; configure managed identities, RBAC, and private networking; deploy with Bicep and the Azure CLI.
• Deploy foundation models with serverless API endpoints or managed compute, choose the right model for a use case, and know when provisioned throughput units make sense.
• Create prompt variants, compare them, and keep prompts under version control in Git.
• Modules: Plan and prepare a GenAIOps solution and Manage prompts for agents in Microsoft Foundry with GitHub.

Week 5: Evaluation, observability, and optimization (two domains, 10–15% each)

• Build test datasets, run quality evaluations (groundedness, relevance, coherence, fluency) and risk and safety evaluations, then automate them.
• Monitor latency, throughput, and token cost, and set up logging and tracing.
• Tune retrieval-augmented generation (RAG): chunk sizes, similarity thresholds, hybrid search, and embedding model choice.
• Understand fine-tuning with synthetic data, from development to production.
• Modules: Evaluate and optimize AI agents through structured experiments, Automate AI evaluations with Microsoft Foundry and GitHub Actions, Monitor your generative AI application, and Analyze and debug your generative AI app with tracing.

Week 6: Practice, gaps, and booking

• Take the official practice assessment. Write down every topic you missed, not just your score.
• Go back through the study guide and mark each bullet green, yellow, or red. Spend the week on the reds.
• Redo two labs from memory, without the instructions open.
• Walk through the exam sandbox once.
• Book the exam when you can explain every bullet out loud. Microsoft lets you schedule up to 90 days ahead.

Hands-on lab checklist

Microsoft’s own exam FAQ says hands-on experience with the technology is required to pass its certification exams. These are the ones most worth building yourself:

1. Deploy an Azure Machine Learning workspace with Bicep or the Azure CLI, delete it, and redeploy it from your template.
2. Log two training runs with MLflow and compare their metrics.
3. Run a hyperparameter sweep and pick the best trial.
4. Build a two-step pipeline from components and put it on a schedule.
5. Trigger a training job from a GitHub Actions workflow.
6. Register an MLflow model and deploy it to a managed online endpoint, then to a batch endpoint.
7. Add a second deployment behind the same online endpoint, send it a small share of traffic, then roll back.
8. Create a Foundry project, deploy a foundation model, and keep its prompt in a Git repo.
9. Run a Foundry evaluation for groundedness and safety, then run the same evaluation from GitHub Actions.
10. Turn on tracing for a generative AI app and find the slowest step and the token cost of one request.

A study prompt that turns any skill into practice

An AI chatbot makes a good study partner if you keep it tied to the official study guide. Paste in one bullet at a time:

You are an Azure MLOps engineer and exam coach. I'm studying for Microsoft exam AI-300.

Skill from the official study guide:
[PASTE ONE STUDY GUIDE BULLET HERE]

1. Explain the skill in plain English in under 100 words.
2. Show how I'd do it with the Azure CLI or the Azure Machine Learning Python SDK v2. Flag any step you're not sure is current.
3. Write 3 scenario questions that make me choose between two similar options (for example, a real-time vs. a batch endpoint). Hold the answers.
4. After I reply, grade me and name the Microsoft Learn page that confirms each answer.

Rules: stick to generally available features. Use current names (Microsoft Foundry, not Azure AI Foundry). If you don't know, say so.

You are an Azure MLOps engineer and exam coach. I'm studying for Microsoft exam AI-300.

Skill from the official study guide:
[PASTE ONE STUDY GUIDE BULLET HERE]

1. Explain the skill in plain English in under 100 words.
2. Show how I'd do it with the Azure CLI or the Azure Machine Learning Python SDK v2. Flag any step you're not sure is current.
3. Write 3 scenario questions that make me choose between two similar options (for example, a real-time vs. a batch endpoint). Hold the answers.
4. After I reply, grade me and name the Microsoft Learn page that confirms each answer.

Rules: stick to generally available features. Use current names (Microsoft Foundry, not Azure AI Foundry). If you don't know, say so.

AI tools can lag behind Azure’s frequent renames and updates, so check anything important against Microsoft Learn before you memorize it. Our 3-minute fact-check workflow makes that quick.

Exam-day tips

• Don’t expect a published question format. Microsoft doesn’t reveal question types before the exam. Try the free exam sandbox to learn the interface, and read the overview screens at the start of your exam.
• You can look things up on Microsoft Learn. Associate exams let you open Microsoft Learn in a split screen. The clock keeps running, so use it to confirm a CLI parameter, not to learn a topic.
• Breaks cost you questions. Five minutes of break time are built in, but the clock keeps running and you can’t return to questions you viewed before a break. Details are on Microsoft’s exam duration and experience page.
• It’s English only. If the exam isn’t in your preferred language, the study guide says you can request an additional 30 minutes. Do it before you book.
• Know the current names. Azure AI Foundry is now Microsoft Foundry, so older tutorials use the old name. Most questions cover generally available features, though commonly used previews can appear.
• Register with a personal Microsoft account. Microsoft warns that exam records tied to a work or school account are lost if you leave that organization.
• If you fail, regroup fast. You can retake 24 hours after a first attempt.

Common mistakes to avoid

• Studying only from DP-100 material. Your Azure Machine Learning skills carry over, but the GenAIOps domains and the DevOps tooling are new.
• Treating the generative AI half as prompt writing. The exam is about operating these systems: evaluation, monitoring, cost, throughput, and safe deployment.

After you pass: renewal

AI-300 is an associate certification, so it expires one year after you earn it. Renewal is free: in the six months before expiry, you pass a shorter, unproctored, open-book assessment on Microsoft Learn. Set a calendar reminder for month six. Our guide to AI certification expiry rules compares this with AWS, Google, and NVIDIA.

Try it now

Open the AI-300 study guide, paste the skill bullets into a document, and mark each one green, yellow, or red. Then put an exam date six weeks out on your calendar. Still choosing between Microsoft and AWS for MLOps? Read AI-300 vs the AWS Machine Learning Engineer Associate first. Free practice questions for other AI exams are on our AI exam prep hub.

Related guides

• Microsoft AI certifications in 2026: the new path, explained
• How to pass AI-103: a 3-week study plan
• Is the AI-300 Certification Worth It? An Honest Look at Microsoft’s MLOps Cert
• AI-300 vs AWS ML Engineer Associate: Which MLOps Cert?
• DP-100 Is Retired: What Replaced Azure’s Data Science Certification

HOW TO // AI is not affiliated with or endorsed by Microsoft. Exam details come from Microsoft’s official pages as of September 2026 and can change. Check the official AI-300 exam page before you book.