Azure AI and Machine Learning Services

Quick Answer: Azure AI Services = ready-made vision/speech/language models you call by API. Azure Machine Learning = platform to build your OWN models. Azure OpenAI Service = GPT-4o, DALL-E, Whisper inside Azure. Azure AI Foundry = the portal that ties generative-AI building together. Azure Bot Service = conversational bots.

Pre-built vs. Custom: the core distinction

The single most-tested idea here is pre-built versus custom. If a scenario says the team has no data scientists or wants results out of the box, the answer is Azure AI Services. If it says they must train a model on their own labeled data, the answer is Azure Machine Learning. Microsoft renamed Cognitive Services to Azure AI Services in 2023; both names can appear, and they mean the same thing.

Azure AI Services (formerly Cognitive Services)

Azure AI Services exposes pre-trained models through REST APIs and SDKs. You send an image, audio clip, or text and get a structured result back — sentiment, transcribed words, detected objects. They group into four families:

Trap: Custom Vision is part of AI Services even though you upload your own images — it is a guided, low-effort trainer, not the full Machine Learning platform. If the question stresses minimal effort and image tagging, choose Custom Vision; if it stresses full control over algorithms and pipelines, choose Azure Machine Learning.

Azure Machine Learning

Azure Machine Learning is the end-to-end platform for the custom-model lifecycle: prepare data, train, evaluate, deploy as an endpoint, and monitor. Know these capabilities by name:

• Automated ML (AutoML) — tries multiple algorithms and tunes hyperparameters automatically; the answer when the requirement is build a model with limited data-science skill but still custom.
• Designer — drag-and-drop visual pipeline builder (no/low code).
• Notebooks — code-first Jupyter for data scientists.
• MLOps — CI/CD, versioning, and the model registry for production models.

Worked example: A retailer has three years of labeled sales data and wants to forecast demand per store. There is no off-the-shelf API for their numbers, so the answer is Azure Machine Learning (likely AutoML for the forecasting task), not AI Services.

Azure OpenAI Service

Azure OpenAI Service delivers OpenAI's GPT-4o, GPT-4, DALL-E (images), and Whisper (transcription) models, but running inside Azure's network, role-based access control, and compliance certifications. The exam angle: choose Azure OpenAI over the public ChatGPT API when the requirement names enterprise security, data residency, private networking, or regional compliance. Typical uses: summarization, content and code generation, chat assistants, and image generation.

Azure AI Foundry, Copilot, and Responsible AI

The January 14, 2026 objective refresh expanded generative-AI coverage. Know three terms at a glance:

• Azure AI Foundry (formerly Azure AI Studio) — the single portal to build, test, and deploy generative-AI apps and agents.
• Microsoft Copilot — AI assistants embedded across Microsoft 365, Security, and the Azure portal.
• Responsible AI — Microsoft's six principles: fairness, reliability & safety, privacy & security, inclusiveness, transparency, and accountability.

Azure Bot Service

Azure Bot Service hosts conversational bots and connects them to many channels (Microsoft Teams, web chat, Slack, Facebook). It plugs into AI Language for natural-language understanding. Build code-first with the Bot Framework SDK or no-code with Microsoft Copilot Studio (formerly Power Virtual Agents).

How the AI services relate to each other

New learners often blur these five offerings, so it helps to see how they stack. Azure AI Services sits at the application layer — you consume finished intelligence with a few API calls and never see a training loop. Azure Machine Learning sits one layer down at the platform level, where you actually create the intelligence from raw data. Azure OpenAI Service is a specialized slice of generative models that you also consume by API, but with the added enterprise guarantees a regulated customer needs.

Azure AI Foundry is the workbench that ties prompts, models, and evaluation together for generative apps, and Azure Bot Service is a delivery channel that surfaces any of this intelligence to end users in chat.

A useful mental rule for the exam: if the requirement could be satisfied by calling an API with no training, it is AI Services or Azure OpenAI; if it requires creating something new from the customer's own data, it is Azure Machine Learning. Watch for the word custom — it almost always redirects you away from pre-built AI Services toward Machine Learning, with the lone exception of Custom Vision, which is a guided trainer that still lives inside AI Services.

Also remember that Responsible AI is a set of governance principles, not a product you deploy, so a question asking which Microsoft framework promotes fairness and transparency in AI is testing the principles, not a service name.

On the Exam: Map keywords, not architecture. Pre-built / no ML skills → Azure AI Services. Train custom model → Azure Machine Learning. GPT / DALL-E in a compliant boundary → Azure OpenAI Service. Chatbot across channels → Azure Bot Service. You will never write code or train a model on this exam.