Copilot Studio in AB-730 Explained: Building Custom Agents for the AI Business Professional Exam

Last Updated: May 2026

TL;DR: The AB-730 study guide never names Copilot Studio, yet five agent creation bullets in the highest weighted exam domain map directly to it. This guide walks UK candidates through what Copilot Studio is, how declarative and custom engine agents differ, and the build, publish, and governance flows AB-730 tests [1] [3].

Copilot Studio sits behind almost every agent question on the Microsoft Certified: AI Business Professional (AB-730) certification, yet you can read the official study guide twice and never see the words "Copilot Studio" appear in print. That gap catches candidates out. The exam tests whether you can create, configure, and share Microsoft 365 Copilot agents, and Copilot Studio is the low-code platform that makes those skills tangible. Understanding the tool, even at a business user level, is one of the highest leverage moves you can make for AB-730 preparation.

This guide maps Copilot Studio to the AB-730 skills outline domain by domain, walks through a build, and covers the governance controls Microsoft expects business professionals to recognise. For a wider view of the certification, start with our complete AB-730 study guide, and if you have not yet booked the exam, the AB-730 exam page summarises the format and pricing.

What Copilot Studio Is and Why AB-730 Tests It

Copilot Studio is Microsoft's low-code platform for building agents that extend Microsoft 365 Copilot. Microsoft's own definition is precise:

Copilot Studio is a graphical, low-code tool for building agents and agent flows.

— Microsoft Learn, Copilot Studio fundamentals, updated 8 May 2026 [3]

You reach it at copilotstudio.microsoft.com. The maker portal lets you describe an agent in plain English, then point it at knowledge sources, configure how it answers, and publish it to Microsoft 365 Copilot, Teams, or external channels. No code is required for most scenarios, although developers can add custom topics, plugins, and pro-code extensions through Microsoft 365 Agents Toolkit when low-code falls short.

The AB-730 study guide does not list Copilot Studio by name, but four of its five "Create and manage Microsoft 365 Copilot agents" bullets describe exactly what Copilot Studio produces [1]:

• Understand when to use Agent Store versus creating a new agent
• Create an agent by using a template
• Configure an agent that has knowledge
• Configure agent settings such as instructions, capabilities and suggested prompts
• Share an agent with team members

Those bullets sit inside the "Manage prompts and conversations by using AI" domain, the highest weighted domain on AB-730 at 35 to 40 per cent of the exam [1]. In practice, expect three to five agent related questions on a 40 to 60 question paper. Microsoft tests these skills through both the in-product Agent Builder experience inside Microsoft 365 Copilot Chat and through the Copilot Studio maker portal. The two share concepts and terminology, so understanding Copilot Studio gives you the conceptual model you need either way.

A brief historical note: Copilot Studio was rebranded from Power Virtual Agents at Microsoft Ignite in November 2023. If you come across older revision guides that mention PVA, treat the underlying product knowledge as still useful but check terminology against current Microsoft Learn pages. For a business level overview of what agents can actually do for an organisation, see our business overview of Copilot agents.

Declarative Agents vs Custom Engine Agents: What AB-730 Tests

The AB-730 study guide bullet "Understand the use case for creating your own agent" assumes you can choose between agent types when the scenario calls for it. Microsoft distinguishes two architectural categories: declarative agents and custom engine agents [5] [6].

A declarative agent runs on the same orchestrator and foundation models that power Microsoft 365 Copilot. You provide the instructions, knowledge, and actions; Microsoft hosts everything else. The Microsoft Learn definition makes the trade-off explicit:

Declarative agents enable you to customize Microsoft 365 Copilot to help you meet the unique business needs of your users... Declarative agents run on the same orchestrator, foundation models, and trusted AI services that power Microsoft 365 Copilot.

— Microsoft Learn, Declarative agents for Microsoft 365 Copilot, updated 24 March 2026 [5]

A custom engine agent gives developers the flexibility to bring their own orchestration and AI services. That means Semantic Kernel, LangChain, or another framework; fine-tuned or open-weight models; and hosting outside Microsoft 365. The trade-off is full control in exchange for separate hosting costs and a more complex deployment.

For AB-730 you do not need to write code for either type, but you do need to recognise which fits a given business scenario. The exam is fond of "company X wants to do Y" stems. A useful rule of thumb:

When in doubt, Microsoft's own example for an employee onboarding assistant points to Copilot Studio with declarative agents:

Use Copilot Studio because it enables rapid, low-code development, and deployment.

— Microsoft Learn, Custom engine agents overview [6]

The study guide does not test declarative versus custom engine by name, so do not memorise terminology for its own sake. Memorise the trade-offs. Declarative agents are the default answer when a question describes a Microsoft 365 grounded scenario without external orchestration requirements.

Building Custom Agents in Copilot Studio: AB-730 Walkthrough

If you have a Microsoft 365 Copilot licence in your tenant, Copilot Studio is included at no additional cost for building agents that run inside Microsoft 365 [10]. The maker portal entry point is copilotstudio.microsoft.com. The build flow Microsoft documents looks like this [3]:

1. Sign in to copilotstudio.microsoft.com with a work or school account.
2. Select Create, then choose New agent, a template, or Describe to design for natural language agent creation.
3. Describe the agent's purpose in plain English. Copilot Studio drafts a starting configuration.
4. Add knowledge sources (SharePoint, uploaded documents, public websites, Dataverse, or enterprise connectors).
5. Refine instructions, suggested prompts, and capabilities in the agent settings.
6. Test the agent in the side test panel.
7. Publish across channels when ready.

The AB-730 bullet "Create an agent by using a template" maps to step 2. Templates exist for scenarios such as a website Q and A bot, an FAQ knowledge agent, and a procedural walkthrough agent. Microsoft scopes templates as opinionated starting points, not finished solutions, so candidates are expected to recognise that a templated agent still requires knowledge and instruction configuration before it is useful.

The "Configure an agent that has knowledge" bullet maps to step 4. Knowledge is what turns a generic chat experience into an agent answering from your organisation's content. Copilot Studio supports five knowledge source categories, summarised in the next section.

The "Configure agent settings such as instructions, capabilities and suggested prompts" bullet maps to step 5. Three concepts matter here:

• Instructions are the natural language guidance you give the agent about its purpose, tone, and constraints. Think of them as a system prompt expressed in a sentence or two.
• Capabilities are toggles such as image generation, code interpreter, web search, or specific actions exposed through plugins.
• Suggested prompts are the example questions a user sees when they open the agent in Microsoft 365 Copilot. They are the equivalent of placeholder text in a search box.

The exam may also reference Topics, which are conversational threads that handle specific user intents. Microsoft defines a topic as "a portion of a conversational thread between a user and the agent" [3]. A topic contains a trigger phrase, a flow of message and question nodes, and optionally branching logic. The default Conversational boosting topic is created for you and handles generative answers from knowledge sources. Business level candidates rarely need to author custom topics, but you should recognise the term if it appears in a scenario.

Configuring Agent Skills, Topics and Knowledge Sources

Knowledge is the heart of an agent. Copilot Studio's knowledge sources are summarised on Microsoft Learn as five categories [4]:

For business users preparing for AB-730, three details are worth memorising.

First, knowledge limits depend on orchestration mode. In generative orchestration mode the agent can use up to 25 public websites, up to 25 SharePoint URLs, unlimited Dataverse tables, and unlimited connector sources. In classic orchestration mode the limits are tighter: four SharePoint URLs, four website URLs, and two Dataverse sources with up to 15 tables each [4]. The exam may pose scenarios about how much content an agent can ground on, particularly when an organisation has a content estate larger than the classic limits.

Second, SharePoint and Microsoft 365 grounding behave best when the tenant has at least one Microsoft 365 Copilot licence. That enables the semantic index (also called Work IQ), which improves how relevance is calculated across SharePoint content. Without it, retrieval falls back to standard SharePoint search.

Third, file size limits matter when uploading documents. With Work IQ enabled, files up to 200 MB are supported. SharePoint and connector files up to 512 MB are supported for common formats including PDF, PPTX, and DOCX [4]. Memorise the ranges, not the exact figures, since Microsoft adjusts these periodically.

Suggested prompts deserve a brief callout. The five questions that appear when a user first opens an agent strongly influence how the agent gets used. A finance assistant that surfaces "What is our quarterly forecast?" will attract very different questions to one that surfaces "How do I submit a purchase order?" The AB-730 audience is expected to understand that suggested prompts shape behaviour even though they look cosmetic. They are part of how a business user controls the agent's perceived scope without writing rules.

Prompt Design Inside Copilot Studio (Different from M365 Chat)

Prompt design inside Copilot Studio is not the same skill as writing prompts in Microsoft 365 Copilot Chat. The end user prompt skills tested elsewhere in the exam (covered in our Copilot prompt engineering techniques guide) focus on what a knowledge worker types into the chat box. Prompt design inside Copilot Studio is what an agent maker writes once, at build time, to control how the agent responds to thousands of subsequent user prompts.

Two artefacts dominate prompt design in Copilot Studio:

The agent instruction is a single block of natural language guidance, usually a paragraph, that defines the agent's role, tone, scope, and refusal behaviour. A well written instruction might read: "You are the HR Assistant for Examinotion Ltd. Answer questions about UK employment policies grounded in the linked SharePoint documents. Refuse to advise on individual disciplinary cases and direct users to HR Business Partners. Use British English throughout."

The suggested prompts are five short example questions that appear in the agent's launch screen in Microsoft 365 Copilot. They prime user expectations and channel queries toward what the agent can actually answer.

The AB-730 exam tests both implicitly. Scenario questions describe a business outcome ("the agent should refuse to give legal advice and direct users to the legal team") and ask which configuration achieves it. The answer is almost always the instruction field, not a custom topic. Recognising that distinction saves time in the exam.

A small but useful exam tip: Copilot Studio's instruction field is read by the same orchestrator that powers Microsoft 365 Copilot. That means the same prompt patterns that work in Copilot Chat (clear context, specified format, role definition) work in agent instructions. Treat the instruction field as a one-shot system prompt the agent will follow on every turn.

Publishing and Deploying Agents to Microsoft 365 Copilot

Publishing is where agent making meets governance. AB-730 tests the "Share an agent with team members" bullet, which sounds simple but covers a tenant-wide approval flow [9]. The Microsoft 365 admin centre documents the workflow as five steps:

1. Maker builds the agent in Copilot Studio and selects Publish.
2. The agent appears in the Requests tab under Agents > All agents in the Microsoft 365 admin centre.
3. An admin reviews the agent's capabilities, knowledge sources, and maker details.
4. The admin selects Publish to add the agent to the "Built by your org" collection in the Agent Store, or Reject.
5. On publishing, the admin selects scope: everyone in the organisation, no one, or specific users or groups.

The role required is AI Admin, not Global Administrator. Microsoft is explicit about this:

The following administrator roles can manage agents in the Microsoft 365 admin center: AI Admin; Global Reader (view-only, no edit). Use roles with the fewest permissions... Global Administrator is a highly privileged role. Limit its use to emergency scenarios.

— Microsoft Learn, Manage agents in the Microsoft 365 admin centre, updated 6 May 2026 [9]

AB-730 candidates should remember three things here. The AI Admin role exists and is the right least-privilege role for agent management. The publish-to-tenant workflow goes through an admin approval gate, not directly from maker to user. And agents can be scoped after approval to specific users or groups, not only "everyone" or "no one".

Agents are categorised in the admin centre by source: published by your organisation, shared by creator (informal sharing without approval), Microsoft agents, external partner agents, and frontier agents such as the App Builder agent and Workflows agent. The exam occasionally tests recognition of these categories in scenario questions about who can install what.

For a quick check after publishing, agents go live for users through the Microsoft 365 Copilot Agent Store within minutes. Users find them under "Agents" inside Microsoft 365 Copilot, Teams, or the Microsoft 365 mobile app, depending on the channels chosen at publish time.

Governance, DLP and Environment Controls (Exam-Critical)

Governance is where AB-730 separates business professionals who can build agents from those who can build agents responsibly. The exam tests recognition of governance controls under the same agent creation domain. Two control surfaces matter most: data loss prevention (DLP) policies and environment level controls.

DLP enforcement is no longer optional. Microsoft confirmed in an early 2025 message centre alert that DLP applies to all tenants:

Since early 2025, data policy enforcement is in effect for all tenants, as announced in the message center alert MC973179: Copilot Studio - Upcoming updates to data loss prevention enforcement. Agent data policy enforcement exemption is no longer supported.

— Microsoft Learn, Configure data policies for agents, updated 24 April 2026 [7]

Admins configure DLP in the Power Platform admin centre by classifying connectors as Business, Non-business, or Blocked. The exam is unlikely to test specific connector names, but candidates should recognise what DLP can block in Copilot Studio. The granular controls include [7]:

• Requiring Microsoft Entra ID authentication for chat with users
• Blocking specific knowledge source types (SharePoint, public websites, uploaded documents)
• Blocking HTTP connector requests from agents
• Blocking skills from being added to agents
• Blocking publishing to specific channels such as Teams or the M365 channel
• Blocking event triggers

DLP enforcement is real time. Agent makers see error messages when they try to add a blocked source, and users see error messages when an agent attempts a blocked action. Policies can be set at tenant level (applying to all environments) or at individual environment level for finer control.

The wider security and governance feature set includes sensitivity labels, customer-managed encryption keys, audit logs via Microsoft Purview and Microsoft Sentinel, Customer Lockbox support, and runtime protection status visible to agent makers [8]. AB-730 candidates do not need to configure these, but they should recognise that Microsoft Purview is the audit system of record for Copilot Studio activity. Memorise the common AB-730 exam mistakes post if you tend to confuse Purview with Defender on certification questions.

For UK organisations the data residency story is meaningful. EU Data Boundary compliance is available for tenants billed in the EU or EFTA whose environments sit inside the EU geographic region [14]. Separately, the United Kingdom was one of the first four countries (alongside Australia, India, and Japan) to receive in-country data processing for Microsoft 365 Copilot interactions by the end of 2025 [11]. Microsoft positioned this as enabling government and highly regulated industries to access Copilot with stronger governance.

That regulatory groundwork has not been theoretical. A GOV.UK report on a cross-government Copilot trial across 20,000 civil servants in 12 organisations found average time savings of 26 minutes per day and a satisfaction score of 7.7 out of 10. The report noted that "Copilot Agents were introduced mid-trial, generating strong positive feedback" [12]. The trial covered the consumer experience rather than maker-side Copilot Studio specifically, but it tells you the policy direction the UK public sector is moving in.

Practice Scenarios: How AB-730 Questions Frame Copilot Studio

The most useful preparation you can do is practise reading exam-style stems and identifying which Copilot Studio capability the question is testing. Three stylised examples follow. None of these are real exam questions and reproducing real exam content is both copyright infringement and a violation of the candidate non-disclosure agreement.

Scenario one. A finance team wants an agent that answers questions about expense policy. The policy lives in 18 SharePoint files. The team has Microsoft 365 Copilot licences. Which agent configuration meets the requirement with the least configuration effort? The exam-correct answer points to a declarative agent built in Copilot Studio with SharePoint as the knowledge source. No external orchestration is needed; the data lives inside Microsoft 365; and the licence requirement is already met for semantic index.

Scenario two. A retail organisation wants an agent its store managers can use in Teams to ask about staffing rules grounded on internal documents. They also want to block the agent from answering questions on social media topics. The recognise-and-pick answer is twofold: build a declarative agent in Copilot Studio, then constrain its scope using the instruction field rather than custom topics. The instruction tells the model what to refuse. Custom topics would be overkill.

Scenario three. A pharmaceutical company needs an agent that integrates with a proprietary clinical decision support model. The model is hosted in Azure OpenAI Service and uses bespoke retrieval logic. The correct answer is a custom engine agent, not a declarative agent, because the orchestration and model selection requirements rule out the Microsoft-hosted defaults.

The AB-730 free practice assessment on Microsoft Learn is the closest you will get to genuine exam wording. Use it. Run through it twice, once before deep revision and once afterwards. For more structured preparation see our 30-day AB-730 study plan, and to compare third-party question banks read our best AB-730 practice tests comparison.

FAQ

Is Copilot Studio on the AB-730 exam?

The phrase "Copilot Studio" does not appear in the AB-730 study guide, but five agent creation bullets in the highest weighted exam domain map directly to its capabilities [1]. Expect three to five agent-related questions on a 40 to 60 question exam where Copilot Studio knowledge gives you the answer faster.

What is the difference between a declarative agent and a custom engine agent?

A declarative agent runs on the same orchestrator and foundation models that power Microsoft 365 Copilot, inheriting M365 hosting and security [5]. A custom engine agent lets developers bring their own orchestration and AI services for full control, but requires separate hosting [6]. Choose declarative for M365 grounded scenarios without external orchestration.

Do I need a Copilot Studio licence to pass AB-730?

No. AB-730 is a recognition exam, not a hands-on lab assessment. You can pass without ever opening Copilot Studio. That said, an hour spent inside the maker portal at copilotstudio.microsoft.com (free trial available) is one of the highest leverage hours of revision time you can spend, because it makes the abstract bullets in the study guide concrete.

How do you publish a Copilot Studio agent to Microsoft 365 Copilot?

The maker selects Publish in Copilot Studio. The agent appears in the Requests tab of the Microsoft 365 admin centre, where an AI Admin reviews capabilities, data access, and maker information, then publishes to everyone, no one, or specific users or groups [9]. Approved agents appear in the Agent Store.

What governance controls does AB-730 expect for Copilot Studio?

Recognise that DLP policies are tenant-wide and mandatory since early 2025 [7], that they can block specific knowledge sources and channels, and that Microsoft Purview is the audit system of record. The AI Admin role manages agents in the admin centre, sensitivity labels propagate from SharePoint into agent responses, and Customer Lockbox support is available for regulated industries [8].

Can you build a Copilot Studio agent without coding for AB-730?

Yes. The AB-730 audience profile is explicit: candidates "take advantage of AI to improve daily work... without building AI apps or writing code" [2]. Copilot Studio's low-code maker portal supports natural language agent creation, templates, and form-based configuration. Coding is only needed for advanced custom engine agent scenarios AB-730 does not test.

Where to go next

Copilot Studio is the most consequential single tool you can understand for AB-730. The exam frames it through business-user scenarios about creating, configuring, sharing, and governing agents, not through configuration screens you need to memorise. Spend an hour inside the maker portal, then revise the study guide bullets with the maker experience fresh in mind. The two reinforce each other.

When you are ready to take the next step, start practising for AB-730 with Examinotion's structured practice tests, pair them with the AB-730 30-day study plan, and review the common AB-730 exam mistakes UK candidates make on agent questions so you do not repeat them on exam day.

Sources

1. Microsoft Learn, Study Guide for AB-730: Microsoft AI Business Professional, updated 30 April 2026, https://learn.microsoft.com/en-us/credentials/certifications/resources/study-guides/ab-730
2. Microsoft Learn, Microsoft Certified: AI Business Professional certification page, updated 4 May 2026, https://learn.microsoft.com/en-us/credentials/certifications/ai-business-professional/
3. Microsoft Learn, What is Copilot Studio, updated 8 May 2026, https://learn.microsoft.com/en-us/microsoft-copilot-studio/fundamentals-what-is-copilot-studio
4. Microsoft Learn, Knowledge sources in Copilot Studio, updated 28 April 2026, https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-copilot-studio
5. Microsoft Learn, Declarative agents for Microsoft 365 Copilot, updated 24 March 2026, https://learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/overview-declarative-agent
6. Microsoft Learn, Custom engine agents for Microsoft 365, updated 24 March 2026, https://learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/overview-custom-engine-agent
7. Microsoft Learn, Configure data policies for agents in Copilot Studio, updated 24 April 2026, https://learn.microsoft.com/en-us/microsoft-copilot-studio/admin-data-loss-prevention
8. Microsoft Learn, Security and governance in Copilot Studio, updated 30 March 2026, https://learn.microsoft.com/en-us/microsoft-copilot-studio/security-and-governance
9. Microsoft Learn, Manage agents in the Microsoft 365 admin centre, updated 6 May 2026, https://learn.microsoft.com/en-us/microsoft-365/admin/manage/manage-copilot-agents-integrated-apps
10. Microsoft Learn, Copilot Studio billing and licensing, updated 28 April 2026, https://learn.microsoft.com/en-us/microsoft-copilot-studio/billing-licensing
11. Microsoft 365 Blog, Microsoft offers in-country data processing to 15 countries, published 4 November 2025 (updated 3 April 2026), https://www.microsoft.com/en-us/microsoft-365/blog/2025/11/04/microsoft-offers-in-country-data-processing-to-15-countries-to-strengthen-sovereign-controls-for-microsoft-365-copilot/
12. GOV.UK, Microsoft 365 Copilot Experiment Cross-Government Findings Report, published 2 June 2025, https://www.gov.uk/government/publications/microsoft-365-copilot-experiment-cross-government-findings-report/microsoft-365-copilot-experiment-cross-government-findings-report-html
13. Microsoft 365 Blog, Microsoft Ignite 2025: Copilot and agents built to power the frontier firm, published 18 November 2025, https://www.microsoft.com/en-us/microsoft-365/blog/2025/11/18/microsoft-ignite-2025-copilot-and-agents-built-to-power-the-frontier-firm/
14. Microsoft Learn, Geographic data residency in Copilot Studio, updated 17 September 2025, https://learn.microsoft.com/en-us/microsoft-copilot-studio/geo-data-residency