Home › Certifications › AI-103 Certification Guide
Microsoft Certification Guide · 2026AI-103 Certification Guide 2026: Azure AI Apps & Agents Developer Associate
Everything about the AI-103 exam in one place — the full syllabus and domain weightings, what changed after the AI-102 retirement, the salary hike Azure AI engineers are commanding in India and the US, and a 12-week roadmap to pass it.
Training & Technology Partners
Microsoft
AI-103 — Azure AI Apps & Agents Developer Associate
The credential this guide covers
Claude
Claude Certified Architect – Foundations (CCAR-F)
The other half of the FDE path — read the CCAR-F guideAI-103 (Developing AI Apps and Agents on Azure) is the exam that earns the Microsoft Certified: Azure AI Apps and Agents Developer Associate credential. It replaced AI-102, which retired on 30 June 2026. It is a 120-minute proctored exam, you need 700 out of 1000 to pass, and it tests whether you can plan, build, secure and monitor generative-AI and agentic solutions on Azure using Python and Microsoft Foundry. It is aimed at working developers, not beginners.
If you have been searching “is AI-103 worth it”, “AI-103 syllabus” or “AI-102 vs AI-103”, this guide answers all three — with the official exam blueprint, real market salary data for India and the US shown side by side, and an honest read on exactly how much of a pay jump the specialisation is actually worth.
AI-103 exam at a glance
| Attribute | Detail |
|---|---|
| Exam code | AI-103 — Developing AI Apps and Agents on Azure |
| Certification earned | Microsoft Certified: Azure AI Apps and Agents Developer Associate |
| Level | Intermediate / Associate |
| Duration | 120 minutes, proctored, may include interactive lab-style components |
| Passing score | 700 out of 1000 |
| Delivery | Pearson VUE — online proctored or at a test centre |
| Exam price | Varies by country/region of proctoring (India pricing is set locally by Microsoft) |
| Languages | English, Chinese (Simplified & Traditional), French, German, Japanese, Korean, Italian, Portuguese (Brazil), Spanish |
| Prerequisites | None enforced — but Python experience and Azure familiarity are assumed |
| Renewal | Expires annually; renew free via an online assessment on Microsoft Learn |
| Retake policy | 24-hour wait after a first fail; longer waits for subsequent attempts |
| Replaces | AI-102 (Azure AI Engineer Associate), retired 30 June 2026 |
What is the AI-103 certification?
The AI-103 certification validates that you can build production AI applications and autonomous agents on Azure — not that you can describe them. Microsoft's own audience profile is blunt about it: you are “an Azure AI engineer who builds, manages, and deploys agents and AI solutions that take advantage of Microsoft Foundry,” and you should already have experience developing apps using Python.
The credential centres on Microsoft Foundry, Microsoft's unified platform for model management, agent configuration and evaluation pipelines. That is the single biggest signal about where this exam sits: it is a build-and-ship certification, tested against the same platform Microsoft expects enterprise teams to standardise on.
AI-102 vs AI-103: what actually changed
This is the highest-volume question on the topic right now, and the answer matters if you have a half-finished AI-102 study plan sitting on your desk.
AI-102 retired on 30 June 2026. AI-103 is its replacement. If you already hold AI-102, your credential remains valid until its expiry and you renew into the current path — but any new candidate should be studying AI-103, not AI-102. Practice material written for AI-102 will actively mislead you on the agentic and Foundry domains.
| Dimension | AI-102 (retired) | AI-103 (current) |
|---|---|---|
| Title | Azure AI Engineer Associate | Azure AI Apps and Agents Developer Associate |
| Centre of gravity | Azure Cognitive Services — discrete AI services stitched together | Microsoft Foundry — unified model, agent and evaluation platform |
| Agents | Barely present | A first-class domain: agent roles, tool schemas, multi-agent orchestration, approval controls |
| Generative AI weight | Secondary | Largest single domain (30–35%) |
| Responsible AI | Light touch | Explicit: guardrails, evaluators, trace logging, provenance, oversight modes |
| Observability | Minimal | Tracing, token analytics, drift, latency breakdowns, grounding quality |
The practical read: AI-103 is a noticeably more engineering-heavy exam. Roughly a third of it is now about building agentic systems, and another quarter is about planning, securing and monitoring them in production. Rote memorisation of service names will not carry you.
AI-103 syllabus and domain weightings
These are the official skills measured weightings from the Microsoft study guide (skills measured as of 16 April 2026). Study time should follow these proportions — this is the single highest-leverage decision in your prep.
AI-103 exam blueprint — share of exam by domain
Official Microsoft weightings. Bars scaled to the largest domain.
Source: Microsoft Learn, “Study guide for Exam AI-103: Developing AI Apps and Agents on Azure”, skills measured as of 16 April 2026.
Generative AI + agents and Plan/Manage together account for roughly 55–65% of the exam. The three “classic AI” domains — vision, text and extraction — share the remaining 30–45%. Candidates who fail AI-103 almost always over-index on the classic domains because that is where the older AI-102 material was strongest.
The five domains as flashcards
Everything the exam can ask you, compressed. Run through these before every mock.
Plan & manage an Azure AI solution
Choosing between LLMs, small language models, multimodal models and Foundry Tools. Retrieval and indexing method selection. Quotas, scaling, rate limits and cost footprints. Managed identity, private networking, keyless credentials, role policies. Safety filters, evaluators, trace logging, provenance metadata and oversight modes.
25–30% of examGenerative AI & agentic solutions
Deploying and consuming models. Implementing RAG in an application. Defining agent roles, goals, conversation tracking and tool schemas. Orchestrated multi-agent solutions with approval controls. Detecting fabrications. Tracing, token analytics, safety signals and latency breakdowns.
30–35% · largest domainComputer vision solutions
Text-to-image and text-to-video generation from prompts and reference media. Inpainting and mask-based edits. Captioning and accessibility alt-text. Azure Content Understanding, single-task and pro mode. Detecting indirect prompt injection embedded in images.
10–15% of examText analysis solutions
Entity, topic and summary extraction into structured JSON. Sentiment, tone and sensitive-content detection. Azure Translator and LLM-powered translation flows. Speech-to-text and text-to-speech as agent modalities, including custom speech models.
10–15% of examInformation extraction
Ingesting and indexing documents, images, audio and video. Semantic, hybrid and vector search for grounding. Enrichment with custom or built-in skills. OCR-driven RAG ingestion. Analyzers producing structured or markdown output for downstream reasoning.
10–15% of examThe bit everyone under-prepares
Cost and quota reasoning, evaluator configuration, and observability. Genuinely examined, thoroughly unglamorous, and exactly where self-taught candidates lose the marks that separate a 690 from a 720.
Spread across domains 1 & 2Is AI-103 worth it?
A certification is worth its cost when it does one of three things: gets you past a screen, justifies a band change, or forces you to learn something you would otherwise skip. AI-103 does all three — with caveats worth stating plainly.
Where it genuinely helps
- Microsoft partner and enterprise hiring. Microsoft partners carry certification quotas tied to their partner tier. A current, named Microsoft credential is a countable asset to those employers in a way a portfolio alone is not.
- Getting past automated screens. “Azure AI Apps and Agents Developer Associate” is an exact-match keyword in ATS filters for Azure-shop AI roles.
- Forcing breadth. Most self-taught GenAI developers are strong on RAG and weak on evaluation, governance, cost control and observability. AI-103 makes you learn the unglamorous 40%.
- Internal mobility and band reviews. For engineers already inside an Azure organisation, it is the cleanest documented case for moving onto the AI platform team — and the cleanest justification for asking to be re-banded.
Where it does not help
- It will not substitute for shipped work. No hiring manager at a serious AI team offers a senior band on a certificate alone. The credential opens the conversation; a deployed system wins it.
- It is not a beginner path. Microsoft assumes you can already write Python and understand cloud application development. Attempting AI-103 without that is the most common reason people fail it twice.
- It expires annually. Renewal is free, but it is a recurring obligation, not a one-time trophy.
What the AI-103 specialisation actually pays
Here is the honest framing before the numbers. No certificate has a salary attached to it. What the market pays for is the role the certificate helps you enter — and that gap is large, widening, and measurable.
What the hike looks like in rupees
The reported specialist premium, applied at its midpoint (~33%) to published Indian cloud and backend developer bands. This is the jump the specialisation buys you.
Entry · 0–2 yrs
Mid · 3–5 yrs
Senior · 6–9 yrs
Lead · 10+ yrs
Azure AI Engineer pay by experience
India and the US are shown as two separate charts, because rupee and dollar bands are different measures on different scales — a shared axis would misrepresent both.
India — annual CTC
Azure AI / GenAI engineer roles, ₹ lakh per annum
Bar length maps the upper bound of each band against a ₹52 L scale. Sources: Glassdoor India AI Architect/Engineer data (July 2026, avg ~₹35 L, range ~₹20.75–47.75 L); published Azure AI engineer India ranges of ₹15–30 L and ₹20–50 L.
United States — annual base
Azure AI engineer roles, US$ thousands
Bar length maps each figure against a $160K scale. Source: published 2026 Azure AI engineer compensation data — average ~$111,000, typical range $90,000–$129,500, top earners around $145,000+.
Generalist vs specialist, side by side
Generalist cloud developer vs Azure AI apps & agents specialist
India, ₹ lakh per annum — modelled from the reported 25–40% specialist premium
Method: generalist bands are drawn from published Indian cloud/backend developer ranges; the specialist bar applies the reported 25–40% AI-specialist premium at its midpoint (~33%). Illustrative modelling, not survey data — treat it as direction, not a quotation.
Compensation figures are compiled from independent, publicly available industry sources and are shown for role context. They are not a guarantee of pay in any specific market, company or outcome, and 360DT does not promise a salary result from any certification or programme.
What is driving AI-103 demand
Certification demand follows platform adoption, and platform adoption follows budget. Three concrete market movements explain why this credential appeared when it did.
1. Microsoft rebuilt the credential around agents, not services
Microsoft did not simply refresh AI-102. It retired it and issued a differently-named certification — Apps and Agents Developer — built on Foundry. Vendors do not rename a certification track casually; the rename is the clearest available signal of where Microsoft expects enterprise AI work to be done for the next several years.
2. The specialist premium is widening, not compressing
Ordinarily, as a skill becomes common, its premium erodes. The reported pattern here is the opposite: the gap between a standard cloud developer and an AI app & agent developer is described as widening through 2026, with AI specialists commanding a 25–40% premium. Independent reporting on agentic AI engineering puts the premium over generalist software engineering at 30–50%, widening to 2–3× for senior multi-agent talent.
3. Agentic AI is the successor wave to “LLM engineer”
The 2023 “LLM engineer” wave had loose scope and looser titles. Agentic AI engineering has firmer technical boundaries — orchestration frameworks, MCP, multi-agent evaluation — and is the function most directly responsible for the AI features end users now interact with. AI-103's largest domain sits exactly on that boundary.
Your 12-week AI-103 roadmap
Study time allocated in proportion to the official domain weightings, the two largest domains front-loaded, and the final fortnight reserved for blueprint-weighted mocks. Assume 8–10 focused hours per week.
Python and Azure baseline
Confirm you are actually ready: Python fundamentals, async calls, REST clients, environment and secret handling, and an Azure subscription with resource groups and RBAC set up. If any of this is shaky, fix it now — not in week 9.
Microsoft Foundry orientation
Create a Foundry project. Deploy an LLM, a small language model and a multimodal model. Learn the model catalogue well enough to justify a choice out loud — the exam tests selection reasoning, not recall.
Domain 2, part one — generative apps and RAG
Build a working RAG application end to end: ingestion, chunking, embeddings, an index, hybrid and vector retrieval, grounded generation. Then break it deliberately and learn what each failure looks like in the traces.
Domain 2, part two — agents and orchestration
Define agent roles, goals and tool schemas. Add function-calling and conversation memory. Build an orchestrated multi-agent workflow with an approval gate. Instrument it and run an error analysis. The single heaviest block of the exam.
Domain 1 — plan, secure, monitor, govern
Managed identity, private networking, keyless credentials, role policies. Quotas, rate limits, cost footprints. Safety filters and content moderation. Evaluators and safety evaluations. Trace logging and provenance metadata. CI/CD for Foundry projects.
Domain 3 — computer vision and multimodal
Image and video generation from prompts and reference media. Inpainting and mask-based edits. Captioning and accessibility alt-text. Content Understanding pipelines, single-task and pro mode. Indirect prompt injection via text embedded in images.
Domains 4–5 — text analysis and extraction
Structured JSON extraction, sentiment and safety detection, Azure Translator, speech as an agent modality. Then ingestion and indexing, enrichment skills, OCR-driven RAG ingestion, and connecting retrieval pipelines to agent tools.
Blueprint-weighted mock exams
Full 120-minute timed mocks weighted to the real blueprint. Review by domain, not by question. Any domain scoring under 70% gets a dedicated re-study day. Run the Microsoft exam sandbox so the interactive components hold no surprises.
Consolidate and sit the exam
Final pass over your weakest two domains, re-read the official study guide bullet list as a checklist, and book the slot. Register with a personal Microsoft account, not a work account — work-account records are lost if you leave the organisation.
Microsoft explicitly recommends registering with a personal MSA account. If you register with an organisational work or school account and later leave that organisation, your exam records are unrecoverable.
Do you need AI-900 first?
No — there is no enforced prerequisite. But the practical path depends on where you are starting.
| Your starting point | Recommended path |
|---|---|
| New to AI and to cloud | AI-900 (fundamentals) → Python for AI → AI-103 |
| Working developer, new to Azure AI | Straight to AI-103, with extra time on Foundry and governance |
| Holds AI-102 | Straight to AI-103 — focus on agents, Foundry, evaluation and observability |
| Data engineer or analyst | Python for AI Developers → RAG & vector DB fundamentals → AI-103 |
Learn from Microsoft Certified Trainers
Certification prep at 360DT is delivered live by Microsoft Certified Trainers, Instructor at 360DT, an Anthropic Partners and a Microsoft Business Applications MVP — our full international faculty, and the same people who run our Forward Deployed Engineer cohorts.
Naga Nandhini
Java Full Stack Corporate Trainer and Enterprise Solutions Expert
Powerhouse corporate trainer with 10+ years of upskilling engineers at HCL, Accenture, Cybage, Mobileum and Renault. An award-winning engineer who helped build the RAPTOR and KIKO platforms for an American Airlines client, she brings real enterprise delivery into every session.
Praveen Kumar
Full Stack and GenAI Corporate Trainer, Ex-TCS Team Lead
Full stack lead and GenAI builder with 10+ years of training behind 50+ corporate batches for TCS, HCL, Perficient, Renault Nissan and Saint-Gobain. He ships real RAG and local-LLM systems with Ollama, FastAPI and Spring Boot, then teaches exactly how they work.
Shantanu Pandey
Microsoft, Google, NVIDIA and HPE Certified Trainer
Director of Engineering at Meteoros Automation, ranked in Microsoft's Top 100 Trainers of 2025, holding MCT, Google Cloud, NVIDIA and PeopleCert credentials, with 500+ trainings delivered since 2011.
Sid
Microsoft Certified Trainer since 2016 and Azure Data Engineer
Dynamic corporate trainer and consultant with more than 10 years of experience across Microsoft Fabric, Azure Data Engineering, and modern cloud platforms. Focused on clear, practical learning that sticks.
Bipeen
AWS, Microsoft, and Anthropic Certified Trainer
AI, machine learning, and cloud transformation expert with more than 25 years of experience. A renowned global corporate trainer and conference speaker known for practical, outcome driven sessions.
Arshad Ahmad
Microsoft Certified Trainer (MCT), Cybersecurity & Power Platform Expert
Technology Trainer with 15 years in strategic client delivery, holding MCT, Azure Solutions Architect Expert, Cybersecurity Architect Expert and Power BI Data Analyst credentials, with 200+ trainings delivered.
Johan
Microsoft Business Applications MVP and Microsoft Certified Trainer
D365 Customer Service Lead – Europe at Avanade, and founder of the Power Platform School. A Microsoft Business Applications MVP and Microsoft Certified Trainer based in London, specializing in Dynamics 365 Customer Service and Power Platform.
Akim Nyamande
Microsoft, CompTIA, Juniper and Cisco Certified Trainer
IT training facilitator with over five years of experience and certifications across Microsoft, CompTIA, Juniper, and Cisco. Previously a Network Administrator before moving into technical training, now delivering hands-on courses in networking, systems administration, and cybersecurity fundamentals.
Ali El Khatib
Cisco Certified Systems Instructor (CCSI) and Microsoft Certified Trainer
Infrastructure Engineer at RHUH with 13 years in the training field, holding CCSI, MCSE, CCNP Routing and Switching, CCNP Security, CompTIA and Microsoft Azure certifications.
Vikas Mittal
AI instructor with 25+ years in technology training
Veteran technology trainer with more than 25 years of rich industry experience spanning Google, Microsoft, and emerging AI platforms. An exceptional mentor recognized for proven teaching excellence.
Sravia
NVIDIA Authorized Instructor and AI Data Specialist
Passionate AI and data evangelist with over 10 years of experience across machine learning, MLOps, and cloud native technologies. Dedicated to delivering impactful, engaging enterprise training programs.
Florian Garcia Compte
Cisco Certified Systems Instructor (CCSI)
Partner Director at AVAANZA FORMACION, a Cisco Learning Partner based in Madrid. Certified Cisco Systems Instructor (CCSI No. 21053) specializing in Cisco Data Center (ACI/SDN), Nexus, and Wireless, delivering official Cisco certification courses including CCNA, CCNP Enterprise, and CCNP Data Center.
AI-103 is one half of the Forward Deployed Engineer path
The 360DT Forward Deployed Engineer Program covers AI-103 and Anthropic's Claude Certified Architect – Foundations, then goes further: RAG, agents, MCP, deployment and the client-facing delivery skills that turn a certified developer into the engineer AI labs compete to hire.
Explore the FDE Program
AI-103 frequently asked questions
Is AI-103 worth it in 2026?
Yes, if you are a Python developer building AI solutions on Azure. It is the current Microsoft credential for the role, it maps to a specialisation reportedly commanding a 25–40% premium over generalist cloud development, and it forces you to learn the governance and observability skills most self-taught GenAI developers skip. It is not worth it as a first certification if you cannot yet write production Python.
How much salary hike can AI-103 give me?
The certificate itself carries no salary. The specialisation does: AI specialists are reported to command a 25–40% premium over standard cloud developers, and agentic AI engineering a 30–50% premium over generalist software engineering. Applied to Indian bands at the midpoint, that is roughly +₹2.5 L at entry, +₹5 L at mid, +₹9 L at senior and +₹12 L at lead level. Those are modelled figures for direction, not a quotation or a guarantee.
Has AI-102 been retired?
Yes. AI-102 retired on 30 June 2026 and AI-103 — Azure AI Apps and Agents Developer Associate — is its replacement. New candidates should study AI-103. AI-102 practice material will mislead you on the agentic, Foundry and responsible-AI domains.
What is the AI-103 passing score?
700 out of 1000. Microsoft uses a scaled score, so it is not a straight 70% of questions answered correctly. You have 120 minutes and the exam is proctored, with possible interactive components.
How long does it take to prepare for AI-103?
For a working developer with Python and some Azure exposure, roughly 10–12 weeks at 8–10 focused hours per week — the roadmap on this page. Candidates coming from a non-development background should add 4–6 weeks of Python and cloud fundamentals first.
Do I need AI-900 before taking AI-103?
There is no enforced prerequisite. AI-900 is useful if you are genuinely new to AI concepts, but a working developer can go straight to AI-103. What you do need is Python experience and familiarity with Azure services — Microsoft states both in the audience profile.
Does AI-103 expire?
Yes. Microsoft associate, expert and specialty certifications expire annually. Renewal is free through an online assessment on Microsoft Learn, taken within the renewal window before expiry.
What is the difference between AI-103 and a Claude certification?
They test different halves of the same job. AI-103 tests building and operating AI systems on Azure infrastructure — deployment, grounding, security, cost, monitoring. Anthropic's Claude Certified Architect – Foundations (CCAR-F) tests agentic architecture, tool and MCP design, and context reliability at the model layer. Holding both is what the Forward Deployed Engineer role actually calls for, which is why our FDE programme prepares for both.
Sources and further reading
- Microsoft Learn — Microsoft Certified: Azure AI Apps and Agents Developer Associate (certification overview, exam duration, languages, retake policy)
- Microsoft Learn — Study guide for Exam AI-103: Developing AI Apps and Agents on Azure (skills measured as of 16 April 2026, domain weightings, renewal, 700 passing score)
- Microsoft Community Hub — announcement of the Azure AI Apps and Agents Developer Associate certification
- Published 2026 Azure AI engineer compensation data — US average ~$111,000, range $90,000–$129,500, top earners ~$145,000+
- Glassdoor India (July 2026) — AI Architect/Engineer average ~₹35 L, typical range ~₹20.75–47.75 L
- Independent 2026 industry reporting on the AI-specialist pay premium (25–40%) and agentic AI engineering premium (30–50%, widening to 2–3× at senior multi-agent level)
360 Digital Transformation is an Anthropic Partner and a Microsoft Authorized Training Partner. Microsoft certification exams are administered by Microsoft through Pearson VUE and are not included in programme tuition. Exam details are accurate as of 20 August 2026; always confirm current format, pricing and skills measured on Microsoft Learn before booking.
Microsoft Foundry
GPT and Foundry Models
Python
