AI-103 Labs: What to Expect and How to Practise in Foundry
Build a small support agent, check what it knows and see when it should stop. Start with the free guided lab, then work towards a project with documents, a tool and acceptance checks.
Start here: your first Foundry support agent
The free AI-103 lab walks through instructions, scope and three checks. Allow 25–40 minutes plus provisioning time; this is a planning estimate, not a measured learner average.
You need your own Azure subscription, a Foundry project, suitable model quota and permission to create and remove practice resources. The guide is free; Azure usage can incur charges. Read the prerequisites before provisioning.
Open the free AI-103 labThree checks before adding more features
The exercise uses a fictional repair business. Its instructions say that repairs are covered for 47 days after collection, exclude liquid damage and give no international shipping price. These facts are supplied in the prompt; this first lab does not use document retrieval.
| Ask in a fresh conversation | Check the actual answer |
|---|---|
| How long are repairs covered, and is liquid damage included? | It should preserve both facts: 47 days and the liquid-damage exclusion. |
| What is the international shipping price? | It should say the price is unknown and suggest human review, without inventing an amount. |
| Write a guide to growing tomatoes. | It should stay within its repair-support scope. |
Record the answer and a pass/fail decision for each check. If a check fails, change one instruction and repeat all three in fresh conversations. A friendly or confident reply does not establish correctness.
Continue with the Complete support project
The introductory lab teaches the basic loop. The Complete project develops one support agent by stages: capture a baseline, ground answers in a policy document, add a constrained support tool, and run acceptance checks. Compare the outputs at each stage so you can explain what the new capability changed.
Compare the current free and Complete packs for question and lab counts, price and access details. The project is a focused part of AI-103 preparation, not coverage of every exam objective.
What we checked, and what you still need to check
In our internal Azure validation on September 16, 2026, all three introductory checks and all six Complete acceptance checks passed in the tested environment. We also inspected baseline and grounded runs and verified removal of the temporary test resources. This is one implementation check, not a learner study, a pass-rate claim or a guarantee for another model or subscription.
Finish by checking your resource inventory. Delete the dedicated lab group only after confirming every resource in it belongs to the exercise, then verify deletion completed. Keep shared resources intact. Closing the browser does not stop Azure charges.
Will AI-103 have a lab?
Treat a lab as possible, not guaranteed. Microsoft does not publish a fixed lab schedule for AI-103 on the current certification or study-guide pages. Candidate reports can describe one appointment, but they cannot predict yours.
Microsoft's official certification exam FAQ is deliberately conditional: if an exam contains labs, its introduction tells you how many labs and tasks are included. Read that introduction before starting and plan your time from the format shown there.
What Microsoft actually confirms
| Confirmed rule | What it means for you |
|---|---|
| Lab information appears in the exam introduction | Do not decide your timing from Reddit reports. Use the item, case-study, lab, and task counts shown in your own appointment. |
| If an exam contains labs, it has at least one | Leave enough time for a performance section when the introduction announces one. |
| Each lab contains between 7–15 tasks | Practise a sequence of small outcomes, not one long scripted deployment. |
| Tasks are scored on the end result | Use the portal, CLI, or PowerShell method you can execute reliably. Microsoft does not require a particular method. |
| You cannot return after leaving a lab | Check every requested outcome before ending the section. |
The Foundry practice loop that matters
The official AI-103 study guide measures five domains. The first two account for 55–65% of the blueprint, so build one small solution that forces you to make the following decisions.
- Create the project deliberately. Choose the resource, project, region, deployment, connection, and model for stated requirements. Record why each choice fits.
- Secure the path. Use managed identity and the narrowest suitable role. Be able to explain private networking, keyless access, quotas, rate limits, and cost controls.
- Ground an application. Ingest a small document set, configure search, retrieve relevant passages, and inspect whether the answer is supported by those passages.
- Add an agent tool. Define a clear tool schema, restrict what it can do, test expected and unexpected inputs, and inspect the trace rather than trusting the final text.
- Evaluate and observe. Compare runs for quality, groundedness, safety, latency, and failures. Change one variable at a time and explain the result.
- Cover the smaller domains. Complete focused exercises for vision, text and speech, and information extraction so you can select the simplest suitable capability.
Microsoft's Foundry training hub links current modules for project setup, model deployment, apps, agents, evaluation, and safeguards. Use those modules to build the workflow, then repeat the critical steps without following the page line by line.
Turn practice into a lab-style drill
Write outcomes, not clicks
Create a short task list such as “deploy a suitable model”, “grant the app access without a key”, or “ground the agent on these files”. Include a way to verify every outcome.
Work from dependencies
Start long-running provisioning first, continue with independent tasks, and avoid changing unrelated defaults. Keep a checklist because a successful click is not proof that the requested state exists.
Inspect the final state
Verify deployments, roles, connections, indexes, agent tools, and test results from the resource itself. Then delete the practice resources you no longer need to control cost.
What this site simulates and what it does not
Examplar provides original scenario practice, explanations and guided lab instructions. You run the labs in your own Azure environment; no hosted sandbox or automatic grading is included. Marking a lab as done records your own check in this browser. These exercises are independent preparation materials, not Microsoft exam tasks or a guarantee about your appointment.
Check the decisions behind your build
Use ten current AI-103 scenarios to find which Foundry choices still need hands-on practice.
Start the free AI-103 diagnosticBuild the rest of the plan
Continue with the 30-day AI-103 study plan, or review what changed from AI-102 to AI-103.