An AI workload is a kind of task that AI solves: understanding text, recognizing speech, describing a photo, reading fields from an invoice. The AI-901 exam wants to know whether you can name the matching workload and the matching service for a scenario. This lesson provides the map. The following lessons of module C1 walk through the areas, and module C2 builds a small helper for each area.
The six workloads
The exam's skills list names six: generative and agentic AI, text analysis, speech, computer vision and information extraction. The course writes the names of the services the way Learn named them at the time of writing.
| Workload | What it does | Service or tool (Learn) |
|---|---|---|
| Generative AI | creates new content such as text, images, audio or code from an input | models in Microsoft Foundry |
| Agentic AI | an agent pursues a goal in several steps and uses tools to do so | Foundry Agent Service |
| Text analysis | pulls features out of text: key phrases, entities, sentiment, summary | Azure Language |
| Speech | turns spoken language into text and text into speech | Azure Speech (in Foundry Tools) |
| Computer vision | describes or analyzes images | models with image input, Azure Vision |
| Information extraction | pulls structured fields from documents, images, audio and video | Azure Content Understanding (in Foundry Tools) |
Foundry Tools is the collective name for these prebuilt services, which you use through a Foundry resource. That is why Learn writes the names as "Azure Language in Foundry Tools" or "Azure Speech in Foundry Tools"; the course shortens them to Azure Language and Azure Speech.
Generative AI
It creates something new instead of merely sorting (lesson 1.2). A chat model writes, summarizes, translates or answers questions. An image model generates images. Fallibility comes with it: you can give tasks to the model, but not responsibility (lesson 1.1).
Agentic AI
A model alone only generates text. According to Learn, it cannot send an email or query a database on its own. An agent adds instructions, a model and tools, such as web search, file search or custom functions. According to Learn, the Foundry Agent Service is a managed platform for building, deploying and scaling such agents. A prompt agent consists of configuration alone (instructions, model, tools) and needs no code of its own. In addition there are voice-based prompt agents and hosted agents with their own code. Lesson 2.4 builds the prompt agent; the AI track shows agents with the Microsoft Agent Framework (AI track, T7.1).
Text analysis
These are fixed tasks on text, often without a chat model. According to Learn, Azure Language offers, among other things, named entity recognition, key phrases, sentiment analysis, summarization, language detection and the detection of personal data. Lesson 1.5 goes into this.
Speech
Azure Speech turns spoken language into text and creates natural-sounding voices from text. Translation and language identification (which language is being spoken) come on top. Also lesson 1.5.
Computer vision
This covers describing and analyzing images, lesson 1.6. As an alternative to the older image analysis service, Learn explicitly names generative models with image input.
Information extraction
It turns unstructured content into structured data. According to Learn, Azure Content Understanding processes documents, images, audio and video into an output format that you define, such as fields with values. Lesson 1.6 names the techniques, and lesson 2.9 builds with it.
Deciding by input and result
A simple rule helps with every assignment. First ask: What goes in? Text, audio, image, document. Then: What should come out? A label (positive or negative), new text, fields with values, an action carried out. The combination usually points to a workload.
The example: Nordlicht makes the assignments
Nordlicht Reisen GmbH has four kinds of sample data and one wish. The table shows how the case assigns them:
| Scenario | Input | Result | Workload |
|---|---|---|---|
| Assess customer reviews (about 300 a day) | text | sentiment, topics | text analysis (Azure Language) |
| Capture hotline calls | audio | text | speech (Azure Speech) |
| Check room photos from travelers | image | description | computer vision (model with image input) |
| Evaluate hotel invoices and confirmations as PDF | document | fields such as amount and date | information extraction (Content Understanding) |
| A helper for customer service that answers questions | text | draft reply, an action if needed | generative and agentic AI (prompt agent reise-helfer) |
This assignment is the case's, not the only possible one. A chat model could also summarize reviews, and according to Learn, Content Understanding offers prebuilt analyses for calls. For each task, the case chooses the service that fits the task most precisely. Module C2 builds them this way.
Real solutions combine workloads. A hotline recording first becomes text (speech) and is then examined for sentiment (text analysis). The agent reise-helfer may call a tool that reads invoice fields (information extraction). Lesson 2.10 puts these decisions together.
Exercise 1
Match each scenario to a workload and name the input and the result:
- A hotel wants to pull the most frequently mentioned topics from guest comments.
- An airline has announcements for its website read aloud in several voices.
- An insurer wants to transfer name, date and amount from scanned claim forms into a database.
- A support team wants an assistant that reads tickets, searches a knowledge collection and creates a draft of the reply.
- A photo agency wants a short description for every uploaded photo.
- A marketing team wants to generate several text variants for an ad from a briefing.
Solution 1
- Text analysis (key phrases). Input text, result a list of topics.
- Speech, more precisely speech output. Input text, result audio.
- Information extraction. Input document, result structured fields.
- Agentic AI. Input text, result a draft; the agent uses tools such as a search.
- Computer vision. Input image, result text. A model with image input can do this.
- Generative AI. Input text, result new text.
For no. 5, an image analysis service would also be conceivable. What counts is the justification through input and result.