MLA-C02 vs AI-103
These two exams are both associate level and both about AI, but they test different jobs. MLA-C02 is about building, deploying and operating machine learning models. AI-103 is about building generative AI applications and agents. Choosing between them on vendor alone misses the more important difference.
Side by side
| MLA-C02 | AI-103 | |
|---|---|---|
| Vendor | AWS | Microsoft |
| Focus | Classical ML engineering and MLOps | Generative AI apps and agents |
| Domains | 4, near-equal weight | 5, front-loaded |
| Heaviest content | Data prep 28%, then three domains at 24% each | Generative AI and agents 30–35%, planning 25–30% |
| Coding | Yes | Yes, Python |
| Production emphasis | Very high | High |
The real difference
MLA-C02 is an operations exam wearing a machine learning badge. Its four domains carry almost equal weight, and three of them — deployment and orchestration, operating and monitoring and securing, plus data preparation — are about getting models into production and keeping them alive. Model development itself is only a quarter.
Expect questions about drift detection, retraining triggers, canary deployments, batch versus real-time serving, training-serving skew and least-privilege roles for training jobs.
AI-103 is a generative AI exam. Over half its weight sits in planning and managing an Azure AI solution plus implementing generative AI and agentic solutions. Vision, text analysis and information extraction are each only 10–15%.
Expect questions about retrieval-augmented generation, chunking strategy, groundedness versus relevance, agent orchestration, prompt injection defence and managing quota and cost.
Which one matches your work
Take MLA-C02 if you train models, own a feature pipeline, run inference endpoints, or get paged when a model degrades. The exam rewards having done this.
Take AI-103 if you build applications on top of foundation models — RAG systems, chatbots grounded in company documents, agents that call tools. The exam rewards having built one and watched it fail in interesting ways.
If you do both, take the one matching where you spend most of your week. The other becomes much easier afterwards because the operational reasoning transfers.
What does not transfer
Less than you might hope, in the middle of each exam.
The MLOps thinking in MLA-C02 — versioning, drift, promotion gates, reproducibility — does transfer to generative AI work, and increasingly people need both. But the specific content does not: knowing how to tune a gradient boosting model helps nothing with chunking strategy, and understanding embeddings helps nothing with class imbalance.
Treat them as two exams, not one exam in two dialects.
Difficulty
Both are genuine associate exams that punish study without practice.
MLA-C02 is arguably the harder of the two for someone without production experience, because three-quarters of it concerns situations you only recognise from having lived them. AI-103 is more approachable for a competent developer who has built a RAG application, but its production and cost management content catches people who have only built prototypes.
Effort
- MLA-C02 — six to eight weeks for a working ML engineer
- AI-103 — four to six weeks for a developer with Azure experience
Both roughly double without the relevant daily work.
What to do next
Take both free practice tests — MLA-C02 and AI-103 — and see which set of questions feels like your job. That is a better signal than any comparison table, including this one.
For what comes before and after each, see the AWS AI path and the Azure AI path.