PL-900 Study Guide

Power Apps is the low-code app-building service in Power Platform. It lets people who are not professional developers build business applications that run on the web and on mobile, using a drag-and-drop designer and a formula language instead of writing traditional code.

What you need to know

• Power Apps builds two classic app types, canvas apps and model-driven apps, plus code apps for pro developers
• Everything starts at make.powerapps.com, and app logic is written in Power Fx, the low-code language shared with Excel
• The business case is speed and cost: business users who understand the process build the app themselves, instead of joining a queue for development time
• Model-driven apps are Dataverse only; canvas apps can reach Dataverse plus hundreds of other sources through connectors
• An app is designated Premium if it uses at least one premium connector, a custom connector, or an on-premises gateway, and its users then need a standalone Power Apps plan

Microsoft Learn resource: What is Power Apps? (opens in a new tab)

Power Automate is the automation service. It replaces the manual, repetitive steps in a business process with flows that run on a trigger, so nobody has to remember to copy the attachment, send the reminder, or re-key the data into the second system.

What you need to know

• Power Automate automates both modern cloud services through connectors and legacy desktop applications through robotic process automation (RPA)
• The value story is time returned to people, fewer human errors, and processes that run the same way every time
• It reaches systems that have no API at all, because desktop flows can drive an application's user interface the way a person would
• Automation is authored by the person who owns the process, which is the same low-code argument as Power Apps
• Power Automate connects to more than 1,400 services, the number Microsoft's own PL-900 training uses

Microsoft Learn resource: What is Power Automate? (opens in a new tab)

Dataverse is the data platform underneath Power Platform. It is not just storage: tables arrive with metadata, relationships, security and business logic already attached, so every app, flow, site and agent that reads the data gets the same rules without anyone rebuilding them.

What you need to know

• Data lives in tables made of rows and columns, with a set of standard tables such as Account, Contact and Case based on the Common Data Model
• Security, relationships and business logic travel with the data, so they apply no matter which app is reading it
• Dynamics 365 applications store their data in Dataverse, so a Power App can build on Dynamics 365 data with no integration layer in between
• Dataverse is the grounding layer for AI: agents in Copilot Studio can use Dataverse data as a knowledge source
• The Microsoft Dataverse connector is premium, and Dataverse storage is metered separately from user licences

Microsoft Learn resource: What is Microsoft Dataverse? (opens in a new tab)

Exam tip: When a scenario says the data must be shared across several apps with one security model, the answer is Dataverse. When it says the data already lives in SharePoint or SQL and must stay there, the answer is a connector.

A connector is a wrapper around a service's API that lets Power Apps, Power Automate, Power Pages and Copilot Studio talk to that service without anybody writing integration code. Connectors are what make the platform useful against systems Microsoft does not own.

What you need to know

• Every connector exposes actions (do something: read, create, update, delete) and, for some services, triggers (start a flow when something happens)
• Connectors are tiered standard or premium; SQL Server, Azure Blob Storage, Dynamics 365 and Dataverse are premium, and premium connectors need a standalone Power Apps or Power Automate plan
• Custom connectors wrap a REST API you define from an OpenAPI (Swagger) definition or a Postman collection, and support OAuth 2.0, Basic and API key authentication
• A custom connector can be shared inside your organisation or submitted to Microsoft for certification so everyone can use it
• Connectors are the unit that data loss prevention (DLP) policies classify, which is how an admin controls where business data is allowed to travel

Microsoft Learn resource: Describe the value of connectors in integrating services and data (opens in a new tab)

Power Pages is the low-code service for building secure, external-facing business websites on top of Dataverse data. Its job is letting people outside your organisation, customers, partners, citizens, see and submit data in a controlled way.

What you need to know

• Power Pages sites read and write the same Dataverse data your apps, flows and agents use, with no copy and no synchronisation job
• Site visitors are stored as contact records in Dataverse, and sign-in can use Microsoft Entra External ID, a Microsoft account or LinkedIn
• Access is controlled by web roles and table permissions; every site has two default web roles, Authenticated Users and Anonymous Users
• Licensing is capacity-based per site per month, bought as authenticated user packs or anonymous user packs, rather than per maker
• Content delivery network and web application firewall protection are included in the licence

Microsoft Learn resource: Power Pages security (opens in a new tab)

Exam tip: Power Pages lost its own exam domain in July 2026, but it did not leave the exam. This one business-value skill in Domain 1 is where it still appears, so learn what the product is for even though you no longer need to know how to build a site.

Generative AI shows up in Power Platform in two distinct places, and the exam cares about the difference. Copilot helps the maker build the solution, while AI Builder and AI prompts put AI inside the finished solution for the people using it.

Copilot in each product: Copilot in Power Apps generates the Dataverse tables and the app from a description, Copilot in Power Automate generates the flow and picks the connectors, Copilot in Power Pages generates site structure and forms, and Copilot in Power BI generates reports from a natural-language question.

AI Builder: the layer that adds AI to the running solution. You either pick a prebuilt model that is ready to use, such as invoice processing, or you train a custom model on your own documents and data, then call it from an app or a flow.

AI prompts: reusable generative components invoked with Power Fx. Prebuilt prompts cover summarise, classify, extract entities, translate and sentiment; custom prompts are solution-aware and can be called from Power Apps formulas, Power Automate, Dataverse low-code plug-ins and Copilot Studio.

What you need to know

• AI Builder and AI prompts run on AI Builder credits, and prompts consume credits every time they run
• Training a model and testing a prompt in prompt builder consume no credits; running a prompt in production does
• Generative features run on the Azure OpenAI Service, and some require an admin to enable data movement across regions
• Admins govern AI prompts per environment, or lock the setting tenant-wide with an environment group rule

Microsoft Learn resource: Describe the value of Copilots and generative AI features in Power Platform (opens in a new tab)

Copilot Studio is the low-code studio for building AI agents. An agent goes further than a chatbot: it can answer from your own content, take action in other systems, and in some configurations run on a schedule or an event with nobody prompting it.

What you need to know

• Copilot Studio lives at copilotstudio.microsoft.com and builds agents, workflows and agent flows
• An agent is extended with topics (authored conversation paths), knowledge (content it answers from) and tools (actions it can take)
• Agents publish to channels: Microsoft Teams, Microsoft 365 Copilot, websites, mobile apps and custom applications
• The business case is deflection and self-service: routine questions get answered without a person, and the agent hands off when it cannot help
• Agents can call other agents, so one agent can front a set of specialised ones

Microsoft Learn resource: Copilot Studio overview (opens in a new tab)

A traditional relational database gives you tables and nothing else. Dataverse gives you tables plus the metadata, security model, relationships and business logic that an application would otherwise have to implement itself, which is why a low-code maker can build against it safely.

What you need to know

• The Common Data Model is the standardised schema underneath, which is why Dynamics 365 data is readable by Power Apps and Power BI with no ETL pipeline
• Dataverse scales without manual index tuning, which a traditional database does not
• Business logic in Dataverse runs on the server, so it applies to every client rather than only the app it was written in

Microsoft Learn resource: Describe differences between traditional databases and Dataverse (opens in a new tab)

Exam tip: Dataverse is not "SQL Server in the cloud". If a question offers that as an option it is wrong; the distinguishing features are metadata, built-in security and server-side logic that travel with the data.

Tables hold rows of data, columns define what each row stores, and relationships connect rows in one table to rows in another. Getting the relationship types right is the highest-value thing you can memorise in this section, because the exam asks about them directly.

What you need to know

• Adding a lookup column automatically creates a one-to-many relationship, and the lookup stores a reference to a row in the other table
• A many-to-many relationship is implemented by an intersect table that Dataverse creates for you
• Common column data types include Text, Number, Date and Time, Choice, Lookup, File, Image and Formula
• Microsoft renamed the terminology in November 2020: entity became table, field became column, record became row, and option set became choice

Microsoft Learn resource: Describe tables, columns, and relationships in Dataverse (opens in a new tab)

Exam tip: One-to-many and many-to-one are the same relationship seen from opposite ends. The only difference that matters is which table gets the lookup column.

Forms control how a user sees and edits a single row. Views control which rows appear in a list and which columns that list shows. Both are configured per table and both travel with the table in a solution.

The four form types:

1. Main: the primary form for viewing and editing a record, organised into tabs and sections, responsive by default
2. Quick Create: a short form optimised for creating a record fast, without opening the full record
3. Quick View: a read-only form that shows data from a related table referenced by a lookup on the current form
4. Card: a compact form used in views through the read-only grid control, designed for mobile

What you need to know

• A table must have at least one main form, and you cannot delete the last one
• Multiple main forms can be assigned to different security roles, with a form order and a fallback form
• Views come in three kinds: Personal (owned by one user, shareable), Public (selectable by everyone in the view selector), and System (Quick Find, Advanced Find, Associated and Lookup, which cannot be deleted and do not appear in the view selector)
• A personal view can be based on a public or system view, but a public view can never be based on a personal one

Microsoft Learn resource: Describe table forms and views (opens in a new tab)

Exam tip: Quick Create is for fast entry of a new record. Quick View is for read-only display of a related record. The names are close and the exam knows it.

Business logic is how you make the data enforce its own rules rather than trusting every app to do it. Dataverse gives you several options, and the exam wants you to pick the lightest one that does the job.

What you need to know

• Business rules use a visual designer with no code and can set a value, make a column required, show or hide it, lock it, or show an error and stop the save
• A formula column uses the fx data type; the formula determines the column's type and the type cannot be changed afterwards
• Power Fx low-code plug-ins run inside the Dataverse transaction, so a failure rolls the whole operation back
• Business process flows are supported only in model-driven apps, and they are now authored through solution explorer rather than in Power Automate

Microsoft Learn resource: Describe the options for business logic in Dataverse (opens in a new tab)

Exam tip: A calculated column computes one row's value in real time. A rollup column aggregates across related rows on a periodic system job, so its value is not instant.

You can now build a Dataverse data model by describing it in plain language. In the Power Apps maker portal this is Tables > New table > Start with Copilot, which opens the data workspace and generates tables, columns, data types, sample data and relationships from your description.

What you need to know

• The Copilot panel offers three modes: Create, Import data and Change, and you keep refining conversationally until you choose Save and exit
• A Table options menu lets you set how many tables to create, how large they should be, and whether to include relationships
• Describing the shape of the data in words, such as "each employee can complete many courses and each course can be completed by many employees", is enough for Copilot to propose the right relationship
• The data workspace does not support every column type: Rich text, Customer, Autonumber and Formula are unsupported
• Copilot is enabled for environments by default, and generally available Copilot features cannot be switched off by an admin without contacting Microsoft Support

Microsoft Learn resource: Create and edit tables, columns, and relationships through conversations (opens in a new tab)

Exam tip: Two Copilot entry points look alike. Tables > New table > Start with Copilot builds the data model. Home > Start with data > Create new data builds the tables and a canvas app on top of them.

An environment is a container for your apps, flows, agents and Dataverse data. It is the security boundary, the data residency boundary and the unit that solutions move between, so almost every governance decision starts with an environment decision.

What you need to know

• Each environment is bound to a region, which is the data residency boundary; the location is chosen at creation for production, sandbox and trial environments
• Environment groups let an admin apply rules to a set of managed environments, and an environment belongs to exactly one group
• Environment routing sends makers to their own personal developer environment instead of the default environment, and it is off by default
• The default environment cannot be manually backed up, and its backups can only be restored to a developer environment

Microsoft Learn resource: Power Platform environments overview (opens in a new tab)

Exam tip: Reset is the sandbox superpower. If a scenario needs an environment wiped back to factory settings, the answer is sandbox (or developer), never production.

Security in Power Platform is layered. Microsoft Entra ID decides who you are, the licence decides what you may use, the environment decides where you can work, Dataverse decides which rows and columns you can see, and DLP policies decide where data is allowed to travel.

What you need to know

• Environment Admin and Environment Maker are the two roles in an environment with no Dataverse database; once Dataverse exists, System Administrator takes over from Environment Admin
• Environment Maker can create apps, flows and connections but has no privileges on data; a maker who needs to change Dataverse tables also needs System Customizer
• Dataverse security is built from business units, security roles and teams; owner teams can own records and hold security roles, while access teams cannot do either and are used for record sharing
• Privileges are cumulative, so the least restrictive assignment wins; you cannot grant organisation-wide read and then hide one row
• Column-level security is configured per column with a column security profile, and it only applies to records the user can already open
• A tenant-level Power Platform administrator does not automatically get Dataverse data access; they need the System Administrator role in that environment

The three DLP connector groups:

1. Business: connectors approved for business data
2. Non-Business: the default group for every connector, including new ones Microsoft adds later
3. Blocked: connectors nobody may use

Connectors from different groups cannot be combined in the same app or flow. Policies are created in the Power Platform admin center under Security, at tenant or environment scope, and where several apply the most restrictive one wins.

Microsoft Learn resource: Describe the Microsoft Power Platform security model (opens in a new tab)

Exam tip: Environment Maker builds things, System Customizer changes the data schema. If the scenario says a maker cannot create a table, the missing role is System Customizer.

Privacy and accessibility are the two commitments Microsoft makes on your behalf and the two things you still have to configure yourself. The platform handles residency, encryption and compliance reporting; you handle the accessibility of what you actually build.

What you need to know

• Customer data stays in the macro region geography tied to your tenant, and choosing the European Union region places data at rest inside the EU Data Boundary
• Data is encrypted by default with Microsoft-managed keys, AES 256-bit at rest and TLS 1.2 or higher in transit; customer-managed keys in Azure Key Vault are a Managed Environments feature
• Microsoft Purview holds the audit and activity logs for Power Apps, Power Automate, Power Pages, Copilot Studio and Dataverse, and is where admins service GDPR data subject requests
• The Accessibility checker lives inside the App checker pane in Power Apps Studio and grades findings as Errors, Warnings and Tips
• Power Platform targets WCAG 2.2 Level AA, and Microsoft publishes per-product Accessibility Conformance Reports covering WCAG, Section 508 and EN 301 549
• Model-driven apps are accessible with no configuration; canvas apps and custom pages are accessible only if the maker makes them so

Microsoft Learn resource: Describe how Microsoft Power Platform supports data privacy and accessibility guidelines (opens in a new tab)

Exam tip: App checker finds formula and runtime problems in a canvas app and hosts the Accessibility checker. Solution checker is a different tool that runs best-practice analysis on a whole solution during ALM.

Monitoring answers two different questions: is the platform healthy, and is anybody using what we built. The Power Platform admin center covers both, and the exam wants you to know which report lives where.

What you need to know

• The Power Platform admin center at admin.powerplatform.microsoft.com is the hub, organised into Manage, Monitor, Security, Actions, Copilot, Deployment and Support
• Tenant-level analytics must be turned on by a tenant admin before tenant-wide reports appear; environment-level analytics are under Manage > Products
• The Monitor area is open to both makers and admins, and its retention is short: event logs 7 days, metrics 28 days
• Application Insights is the deep-telemetry option; it needs an Azure subscription, holds a superset of runtime logs, and lets you set your own retention, but it does not produce recommendations
• Microsoft Purview holds the audit logs, and events surface within 24 hours, so it is not a real-time monitoring tool
• Service health sits in the admin center under Support, while the Message center for upcoming changes is in the Microsoft 365 admin center

Microsoft Learn resource: Describe the monitoring and analytics capabilities of the Microsoft Power Platform (opens in a new tab)

Exam tip: Monitor is built in, free of Azure, short retention, and gives recommendations. Application Insights needs Azure, keeps raw telemetry for as long as you pay for, and gives none. That contrast is the question.

Application lifecycle management is how a solution gets from a maker's development environment to production without anyone exporting a zip file by hand. Pipelines are Microsoft's built-in answer, and they are new enough to the exam that they are worth real study time.

What you need to know

• A pipeline deploys solutions between environments on the classic path of development, test, production, with unmanaged solutions in development and managed solutions everywhere downstream
• A pipeline uses three environment roles: the host (holds all pipeline configuration and run history), development environments and target environments
• All target environments must be Managed Environments; the host and the development environments do not have to be
• The maker runs the pipeline from the unmanaged solution in the development environment; pipelines are not visible from a managed solution or from a target environment
• Stages run in order and the same exported artifact is promoted at each one, so a change cannot skip test and land straight in production
• Pipelines cannot deploy across tenants, which is the point at which Microsoft recommends Azure DevOps or GitHub instead

Microsoft Learn resource: Overview of pipelines in Power Platform (opens in a new tab)

Exam tip: You develop in unmanaged and deploy managed. A pipeline will refuse to deploy an unmanaged solution to a target, and that refusal is the exam question.

A canvas app starts as a blank screen. You drag controls onto it, bind them to data, and write Power Fx formulas to make them behave, which gives you complete control over the layout at the cost of building everything yourself.

What you need to know

• Canvas apps are built in Power Apps Studio, and you choose a Responsive, Tablet or Phone layout when you create one
• They connect to Dataverse plus many other sources through connectors: SharePoint, Excel on OneDrive, SQL Server, Microsoft 365, Dynamics 365, Salesforce and more
• Logic is written in Power Fx; delegable functions such as Filter(), LookUp() and Search() push the work to the data source instead of pulling rows into the client
• The maker needs the Environment Maker security role
• Offline use is available only for Dataverse-backed canvas apps in the Power Apps Mobile player; a canvas app in a browser cannot run offline

Microsoft Learn resource: What are canvas apps? (opens in a new tab)

Exam tip: "Pixel-level control over the interface" and "connects to hundreds of data sources" both mean canvas. "Responsive and accessible with no extra work" means model-driven.

A model-driven app is generated from your data. You define the tables, relationships, forms, views and business processes, and Power Apps builds a consistent interface around them, which suits process-heavy applications where the data model is the point.

What you need to know

• A model-driven app requires a Dataverse database; without a Dataverse data model you cannot build one
• The minimum ingredients are one table, a site map and the app itself, assembled in the app designer by adding pages
• Relationships between tables give automatic navigation between related records, which a canvas app only gets if the maker writes it
• Model-driven apps are automatically responsive and accessible, support multilingual labels, and have native offline capability
• End-user access is controlled by Dataverse security roles, not by app-level sharing alone
• A canvas experience can be embedded inside a model-driven app as a custom page

Microsoft Learn resource: What are model-driven apps in Power Apps? (opens in a new tab)

Plan designer, which Microsoft's documentation now calls plans, starts from a business problem rather than an app. You describe the scenario in plain language and it works out the user roles, the data model and the mix of Power Platform products that would solve it.

What you need to know

• Three AI agents run in sequence: the Requirements Agent generates user roles and user stories, the Data Agent suggests tables, columns and relationships, and the Solution Agent proposes the technology
• The technologies it can propose are canvas apps, model-driven apps, Power Pages sites, Power Automate flows, Power BI reports and Copilot Studio agents
• Entry points are Home > Start with a plan in make.powerapps.com, the Plans item in the left navigation, and Solutions > Create plan from a solution for an existing solution
• Plans is generally available, requires a Dataverse database in the environment, and the result is saved into a solution
• Sharing uses two roles, Viewer and Co-Owner, and only one author can edit a plan at a time

Microsoft Learn resource: Overview of plans (opens in a new tab)

Exam tip: Plan designer starts from a business problem and produces a multi-product architecture. Describing an app to Copilot starts from data and produces one canvas app. If the scenario mentions user roles or user stories, it is Plan designer.

Code apps are the pro-developer option. A developer builds a normal web application in a code editor, then hosts it inside Power Platform so it inherits the platform's identity, connectors and governance instead of needing all of that built from scratch.

What you need to know

• Code apps are built with frameworks such as React, using the Power Apps client library (@microsoft/power-apps) and the Power Apps CLI, and published into an environment with pa app push
• They inherit Microsoft Entra ID authentication, access to Power Platform connectors, DLP policy enforcement at app launch, Conditional Access, app quarantine and tenant isolation
• An admin must enable them per environment under Settings > Product > Features > Power Apps code apps
• End users who run a code app need a Power Apps Premium licence
• Code apps are generally available; Microsoft announced general availability on February 5, 2026

Microsoft Learn resource: Power Apps code apps overview (opens in a new tab)

Exam tip: A code app is a whole application. A code component built with the Power Apps component framework is a single control you drop into a canvas or model-driven app. Both are written in TypeScript and only one is an app.

The supported path is to describe your data rather than your app. From the Power Apps home page you choose Start with data > Create new data, describe what you need to store, and Copilot generates the Dataverse tables and then the canvas app on top of them.

What you need to know

• The Copilot panel offers Create, Import data and Change, so you refine the tables conversationally before the app is generated
• You need a Dataverse database in the environment and the System Customizer security role; without them you are routed to a personal developer environment
• Copilot also helps inside the app: it writes and edits Power Fx formulas in the formula bar and suggests fields
• The feature is region-gated, and depending on where your environment is hosted you may need an admin to enable data movement across regions
• Copilot in Power Apps runs on the Azure OpenAI Service

Microsoft Learn resource: Build apps through conversation with Copilot (opens in a new tab)

The model-driven story is different from the canvas story, and the difference is the exam point. There is no in-browser "describe a whole model-driven app" experience. What you get in the browser is generative pages: one AI-generated page inside an app that already exists.

What you need to know

• A generative page is created in the app designer with Add page > Generative page, then describing the page you want
• The agent generates React code plus business logic, and shows you Preview, Code and Compare views, with an accessibility assistant that can auto-fix findings
• A generative page can reference a maximum of six Dataverse tables
• Generative pages work only in model-driven apps, and they are solution-aware
• A plan can also produce a model-driven app: choosing Create on the model-driven app tile opens the app designer with the tables already added

Microsoft Learn resource: Generate a page using natural language (opens in a new tab)

Exam tip: If a question says "in the browser at make.powerapps.com" and "model-driven", the answer is a generative page, not a whole app. Whole-app generation for model-driven apps runs through external command-line tooling and is still preview.

The vibe experience is Microsoft's newest way to build, and it is genuinely different from everything above: you describe what you want and it generates a modern web app with real source code, along with the data model and the plan behind it.

What you need to know

• It is in preview and lives at vibe.powerapps.com, or through Try new experience (Preview) at the top of Power Apps, which you have to re-enable each time you sign in
• Canvas and model-driven apps are explicitly not supported in the vibe experience; what it produces is generated code
• Plan mode is on by default: the agent proposes a plan and asks clarifying questions before it generates anything
• Generated tables are draft tables held in memory until you explicitly publish them to Dataverse
• A tenant admin must switch on Copilot in Power Apps (preview), it is not available in a default environment, and it is limited to certain regions and to English
• Add work content attaches Word, Excel, email or chat content as business context, and requires a Copilot licence

Microsoft Learn resource: Overview of the new Power Apps vibe experience (preview) (opens in a new tab)

Exam tip: Plans and vibe both take a description and both produce more than an app, so keep them apart by output. Plans is generally available and produces a multi-product solution. Vibe is preview and produces one code-generated web app.

A cloud flow automates services that have an API, through connectors, and runs in the cloud whether anyone is at a desk or not. A desktop flow is robotic process automation: it drives an application's user interface the way a person would, which is how you automate software that offers no other way in.

What you need to know

• The three cloud flow sub-types are automated (an event triggers it), instant (a person starts it) and scheduled (a recurrence trigger runs it)
• Power Automate for desktop is preinstalled on Windows 11
• An attended desktop flow comes with the Power Automate Premium user licence; an unattended one needs the Power Automate Process capacity licence
• A cloud flow calls a desktop flow with the action Run a flow built with Power Automate for desktop, choosing attended or unattended on the action
• Business process flows are no longer created in Power Automate; they are authored through solution explorer and run only in model-driven apps

Microsoft Learn resource: Describe use cases for cloud and desktop flows (opens in a new tab)

Exam tip: Attended RPA is a per-user licence, unattended RPA is a capacity licence tied to a machine. That licensing split is a favourite question.

This skill is a tour of the scenarios Power Automate is used for most. Each one has a signature trigger or action, and knowing the signature is usually enough to answer the question.

Approvals: the action is Start and wait for an approval, and there are four approval types: Approve/Reject first to respond, Approve/Reject everyone must approve, Custom responses wait for one response, and Custom responses wait for all responses. Approvals require Dataverse, and approvers can respond from Outlook, the Teams Approvals app, the Power Automate mobile app or the Approvals hub on the web.

Microsoft Teams: the Teams app is called Workflows. You can start a flow from a message with the For a selected message trigger, and post interactive adaptive cards into a chat or channel.

Outlook: the Office 365 Outlook connector handles work and school accounts; the When a new email arrives (V3) trigger filters on folder, sender, importance, attachments and subject before the flow ever runs.

SharePoint: from a list or library command bar choose Integrate > Power Automate. The common triggers are When an item is created and When an item is created or modified.

Microsoft Forms: the connector has exactly one trigger, When a new response is submitted, and one action, Get response details. The trigger returns only a response ID, so the action is what gives you the answers.

Document automation: AI Builder does the reading. A custom document processing model handles fixed template documents, general documents and invoices, and needs a minimum of five sample documents. The prebuilt invoice model needs no training at all.

What you need to know

• Approvals are stored in Dataverse tables, and the Approvals connector is standard, not premium
• Flows created in the Teams Workflows app always land in the default environment
• The Forms trigger on its own tells you nothing about the answers; you always pair it with Get response details
• AI Builder consumes AI Builder credits at run time

Microsoft Learn resource: Describe Power Automate use cases (opens in a new tab)

Exam tip: Document processing is a custom model you train on at least five samples. Invoice processing is prebuilt and needs no training. If the scenario says "standard supplier invoices, no training data", the answer is the prebuilt model.

Every cloud flow has exactly one trigger and then a series of actions. The trigger decides when the flow runs; the actions decide what it does, and each action's output is available to every action below it.

What you need to know

• The trigger kinds are automated, scheduled (the Recurrence trigger), instant and HTTP (When an HTTP request is received, which creates a callable URL)
• A polling trigger checks the service on an interval; a webhook trigger receives a push notification, which is how most modern Microsoft 365 connectors work
• Control actions shape the logic: Condition (two branches), Switch (many cases), Apply to each, Do until, Parallel branch, Scope and Terminate
• Dynamic content is the output of an earlier step; an expression is a function such as formatDateTime() or concat() that you write yourself
• Flow checker validates the flow before you save, and run history shows what actually happened after
• A connection stores credentials for a non-solution flow; a connection reference is the solution component a solution-aware flow uses so connections can be remapped at import

Microsoft Learn resource: Identify Microsoft Power Automate components (opens in a new tab)

Exam tip: Condition gives you exactly two branches. Switch matches one value against many discrete cases. Reaching for nested Conditions where a Switch belongs is both the design mistake and the exam distractor.

Copilot can build a cloud flow from a sentence. On the Power Automate home page you type what you want into the Create your automation with Copilot box, select Generate, and Copilot proposes a trigger and the actions to go with it.

What you need to know

• After you select Keep it and continue, Copilot checks the connections, and a green tick means the connection is ready while a red mark means it needs attention
• Copilot can set up connections on your behalf and apply parameters it found in your prompt
• Once the flow is created the designer opens with the Copilot pane docked on the right, where you can ask it to change, add or remove actions
• Copilot works in the new designer only, not the classic designer
• Prompts work best in the shape "when X happens, do Y", naming the connector rather than describing it
• Copilot cannot read your run history to debug a failure and cannot query your live data to check a column name

Microsoft Learn resource: Create cloud flows with Copilot (opens in a new tab)

Copilot also sits inside Power Automate for desktop, as a chat panel opened from the top right of the console or the designer. It can generate actions from a description, explain what an existing flow does, summarise a subflow and write scripts.

What you need to know

• Creating a desktop flow from a natural-language description is still preview, and it requires a work or school account
• Copilot can generate only a subset of desktop flow actions, covering things like Excel, files, folders, loops, conditionals, text, variables and scripting
• UI and browser automation actions are not in that subset; where Copilot detects that intent it inserts a recorder placeholder that deliberately errors until you record the steps yourself
• Generated actions appear in the designer as a region bracketed by two automatically generated comments
• Other Copilot skills in Power Automate for desktop include generating a flow description, summarising subflows, and Generate script with Copilot inside the Run PowerShell, Run Python and similar actions

Microsoft Learn resource: Create desktop flows with Copilot (opens in a new tab)

Exam tip: The PL-900 training still teaches Record with Copilot, where you narrate your actions aloud and Copilot builds the flow, but the Power Automate product documentation now marks that feature deprecated and unavailable. Learn it for the exam, and use the ordinary recorder in real work.

An agent in Copilot Studio is an AI assistant you build for a specific job. It holds a conversation, answers from content you point it at, and takes action in other systems, which is what separates it from a search box or a scripted chatbot.

What you need to know

• A conversational agent responds to something a person asks; an autonomous agent is defined by an event trigger and runs with nobody prompting it
• Typical scenarios are customer support and order tracking, employee self-service for HR and IT questions, and monitoring something and acting when a threshold is crossed
• Makers do not have to start from blank: agent templates come preconfigured with instructions, tools and topics
• Agents can hand off to other agents, so one front-door agent can route to specialised ones
• Copilot Studio builds three kinds of thing: agents, workflows and agent flows

Microsoft Learn resource: Agents overview (opens in a new tab)

A topic is one piece of a conversation, built on the authoring canvas out of nodes. Topics are how you take control of an exchange when you need the agent to follow your script rather than improvise.

What you need to know

• The node types are Message, Question, Adaptive Card, Condition, Variable management, Topic management (redirect, transfer, end) and Tool nodes, plus advanced nodes for generative answers and HTTP requests
• Custom topics are the ones you create; system topics are built in and can be edited or switched off but never deleted
• System topics include Conversation Start, Conversational boosting, Escalate, Fallback, Multiple Topics Matched, On Error, Reset Conversation and Sign in
• Under generative orchestration, which is on by default for new agents, the trigger reads "The agent chooses" and the agent picks a topic from its description
• Under classic orchestration the trigger reads "User says a phrase" and matching is driven by trigger phrases; Microsoft recommends 5 to 10 per topic
• Topics accept input and output parameters, and can be edited as YAML in the code editor

Microsoft Learn resource: Create and edit topics (opens in a new tab)

Exam tip: Trigger phrases only matter under classic orchestration. Under generative orchestration the topic description is what the agent reads to decide whether to use it, so a vague description is the bug.

Knowledge is the content an agent answers from. You point the agent at your own material, it retrieves the relevant part at question time, and it cites where the answer came from rather than making something up.

What you need to know

• Supported sources include public websites, uploaded files, SharePoint, OneDrive for Business, Dataverse, Azure AI Search and enterprise systems reached through connectors
• Generative answers is effectively on by default: creating an agent creates the Conversational boosting system topic, and agent-level knowledge sources are added to it
• Allow ungrounded responses is the setting that decides whether the agent may answer from the model alone; turn it off and an unsupported answer triggers Fallback instead
• Documents labelled confidential or highly confidential, and password-protected files, will index but return no answers
• Knowledge grounds an answer; a tool takes an action. Under generative orchestration the agent chooses between them from their names and descriptions

Microsoft Learn resource: Knowledge sources summary (opens in a new tab)

Exam tip: There are two ways to use SharePoint as knowledge. Adding it as a SharePoint knowledge source searches live and honours each user's permissions. Uploading files from SharePoint copies the content into Dataverse, which means it is a snapshot that syncs on a schedule.

A tool is how an agent does something rather than just says something. You add tools on the agent's Tools page, and the agent decides which one to call based on the tool's name and description.

What you need to know

• Tool types are Prompt, Agent flow, Computer use, Custom connector, Model Context Protocol and REST API, plus the prebuilt Power Platform connectors
• MCP (Model Context Protocol) is a standard interface between an AI model and external tools and data. Its value is that the MCP server owns the tool definitions, so a change there reaches every agent without re-authoring any of them
• Generative orchestration must be on to use MCP, and a topic cannot call an MCP server directly
• An agent flow is a deterministic automation authored natively in Copilot Studio; unlike a Power Automate cloud flow it needs no Power Automate licence and consumes Copilot Studio capacity per action
• Tool inputs default to Dynamically fill with AI, so the agent works out the parameter values from the conversation rather than asking a Question node
• Tools run in the user's context, so an agent set to No authentication cannot use tools that need credentials

Microsoft Learn resource: Add tools to custom agents (opens in a new tab)

Exam tip: Converting a Power Automate cloud flow into an agent flow is a one-way operation. There is no route back, which makes it the sort of irreversible detail exams like.

Publishing is what makes an agent real. Until you publish, nobody but you can use it, and you cannot even add a channel. One publish updates the agent everywhere it is connected.

What you need to know

• Channels include Microsoft Teams and Microsoft 365 Copilot, SharePoint, WhatsApp, a demo website, a custom website, a mobile app, Facebook and the Azure Bot Service channels such as Slack, Telegram and email
• Dynamics 365 Customer Service integration is configured separately, under the customer engagement hub
• Authentication gates channels: WhatsApp requires no authentication or manual authentication, and the demo website link only appears when authentication is not set to Microsoft Entra ID
• Choosing No authentication means the agent cannot use tools that run on user credentials
• Publishing in Teams has three steps: install it for yourself, share an install link with colleagues, then Show to the organization, which needs admin approval
• Admins can block channels centrally with Agent access channels in the Power Platform admin center and with DLP connector policies

Microsoft Learn resource: Key concepts - Publish and deploy your agent (opens in a new tab)

Exam tip: The test chat is for the maker while building and does not count in analytics. The demo website is a shareable page for colleagues and stakeholders, and Microsoft says explicitly it is not for production.

Once an organisation has more than a handful of agents, somebody has to know what exists, who owns it and what it can reach. Microsoft Agent 365 is the control plane for that: it extends the tools you already use to manage people so they manage agents too.

What you need to know

• Microsoft frames it around three jobs: observe, govern and secure
• It brings together the agent registry in the Microsoft 365 admin center, Microsoft Entra Agent ID for identity and Conditional Access, Microsoft Purview for compliance and information protection, and Microsoft Defender for threat detection
• Every agent created in Copilot Studio gets an Entra agent identity, and the user who created it is recorded as its sponsor
• The registry surfaces named risks, including shadow agents (no registry entry, no owner or no agent identity), agents with no owner, and agents with excessive permissions
• Entra Agent ID is available to all Microsoft Entra customers, but extending Entra's security features to agents is what Agent 365 adds
• Agent 365 is licensed per user, is included with Microsoft 365 E7, and is an add-on to Microsoft E5, A5 and Business Premium

Microsoft Learn resource: Overview of Microsoft Agent 365 (opens in a new tab)

Exam tip: The Agent 365 registry in the Microsoft 365 admin center covers every agent in the tenant. The Copilot Studio agent inventory in the Power Platform admin center covers only Copilot Studio agents. If a question says "tenant-wide, including agents built outside Power Platform", it is Agent 365.

The Monitor page on an agent answers whether the thing is working. It reports how many conversations happened, how many of those the agent actually handled, and how many ended with a person having to step in.

What you need to know

• Session outcomes are Resolved, Escalated, Abandoned and Unengaged
• Activity from the test panel is not counted, so testing your own agent does not inflate the numbers
• Active users are reported as daily and monthly active users, and this only works when the agent requires authentication
• Conversation transcripts are stored in the Dataverse ConversationTranscript table, which is what you query for custom reporting
• Autonomous agents report run outcomes, trigger use, tool use and knowledge source use rather than conversational metrics

Microsoft Learn resource: Monitor overview (opens in a new tab)

An evaluation is automated, repeatable quality testing for an agent. You write a set of questions, run them all at once, and have the results scored, which is how you check that today's change did not break yesterday's answers.

What you need to know

• The unit of work is a test case, a question with an optional expected response; a group of test cases is a test set
• Test sets can be generated from the agent's own description, from a knowledge source, from a test chat conversation, imported from a CSV, or written by hand
• Scoring is done by test methods, including General quality (relevance, groundedness, completeness and abstention), Content safety, Compare meaning, Tool use, Keyword match, Text similarity and Exact match
• General quality is added to every test set by default and needs no expected answer
• Evaluations can be run from the Copilot Studio interface, the Power Platform REST API, or a connector inside a flow, which is how you put them in a release pipeline
• The test chat is interactive and one question at a time; an evaluation runs a whole test set and is repeatable, so the two are complementary rather than alternatives

Microsoft Learn resource: About agent evaluation (opens in a new tab)

Exam tip: Match the test method to the question. Compare meaning scores intent, Text similarity scores wording, Exact match demands character-for-character identity, and Keyword match only checks that terms appear. Using Keyword match on a question that needs a synthesised answer is the documented wrong choice.