AB-730 Copilot Agents Explained: Agent Store, Agent Builder, Knowledge and Sharing (2026)

Last updated: September 2026. Sources cited throughout are official Microsoft Learn documentation, verified 4 September 2026.

TL;DR: AB-730 tests whether you can choose between installing a prebuilt agent from Agent Store and building one with Agent Builder, configure an agent's knowledge sources and instructions, and share it with colleagues. Agent Store hosts ready-made agents like Facilitator and Planner Agent, while Agent Builder lets you create a custom declarative agent from a template or natural language in minutes. Knowledge sources have precise per-type limits (four public URLs, 100 SharePoint files) that candidates are expected to know.

Microsoft's AB-730: AI Business Professional certification tests practical, day-to-day fluency with Microsoft 365 Copilot, and one of its most heavily weighted objectives is "Create and manage Microsoft 365 Copilot agents" [1]. This single sub-topic sits inside the "Manage prompts and conversations by using AI" domain, which carries 35-40% of the total exam weight, making it one of the highest-value areas to master before exam day.

The skill itself breaks into five testable actions: knowing when to use Agent Store instead of building a new agent, creating an agent from a template, configuring an agent's knowledge, setting an agent's instructions and capabilities, and sharing an agent with team members [1]. This guide walks through each of those five actions using Microsoft's own current documentation, flags the two places where Microsoft's public docs disagree with themselves, and closes with a look at Researcher and Analyst, the two reasoning agents Microsoft added to the AB-730 audience profile in 2026. If you have not yet mapped out how this topic fits into your wider revision, Examinotion's AB-730 30-day study plan shows where Copilot agents sit relative to the exam's other domains.

Agent Store versus building a new agent

Agent Store is the central hub inside Microsoft Copilot where users discover and install agents built by Microsoft, trusted partners, and their own organisation [2]. The exam's first sub-skill, "understand when to use Agent Store versus creating a new agent" [1], is really a question about reuse versus custom-build: if a ready-made agent already does what you need, install it from Agent Store rather than spending time building a near-duplicate.

Agents reach Agent Store through four distinct paths. An administrator can deploy a prebuilt agent that needs no development work at all, a maker can build an agent in Copilot Studio and publish it for admin approval, a developer can bring an agent from an external platform via the Microsoft 365 Agents Toolkit, or an organisation can integrate an agent through the enterprise-grade Agent 365 SDK [2]. Once an admin has made a prebuilt agent available through the Microsoft 365 admin centre, users can find and install it from Agent Store across Teams, Outlook, Word, Excel, and PowerPoint [3]. Before any agent becomes visible to end users this way, it undergoes a streamlined Microsoft submission and approval process, managed by administrators through the Microsoft 365 admin centre [5].

Not every agent reaches a user's toolbar the same way. Microsoft documents three separate installation and governance models [3]:

Several named agents in Agent Store are worth knowing individually for AB-730. Facilitator keeps meetings organised by surfacing agendas, tracking progress, and capturing real-time notes, and it is included with a Copilot licence subject to available capacity [6]. Planner Agent, which Microsoft's current documentation now names rather than the older "Project Manager agent" label, automates plan creation and task execution and offers an in-app chat experience for plan updates and summaries in premium tiers [7]. Interpreter provides real-time speech translation in Teams meetings across nine languages, with 20 hours of interpretation included per user per month on a Copilot licence [8].

Admin controls for these agents vary by product. Facilitator is managed through the Teams admin centre under Manage apps [6], while Planner Agent and Interpreter are both PowerShell-only, configured through Set-PlannerConfiguration and CsTeamsMeetingPolicy respectively [7][8]. Broader tenant-wide agent access, covering which publishers and which users can use agents at all, is set centrally in the Copilot Control System, under Microsoft 365 admin centre → Copilot → Settings → Data access → Agents [4]. Two separate assignment concepts exist there: "Available to" controls who can find and install an agent, while "Deploy to" pushes the agent onto users automatically, accepting the necessary permissions on their behalf [4].

Creating an agent by using a template

The second testable skill, "create an agent by using a template" [1], is handled by Agent Builder, a tool available from microsoft365.com/chat, office.com/chat, and the Teams desktop and web clients (not on mobile) [9]. Agent Builder is deliberately lightweight: Microsoft's own guidance positions it as ideal "for quick and straightforward projects" [9], reserving Copilot Studio for agents that need external service integrations.

One correction worth making early: despite the exam objective's phrasing sitting inside a broadly "Microsoft 365" context, Agent Builder itself does not launch from SharePoint. SharePoint has a separate, simpler agent-creation tool for building agents scoped to a single site, and those SharePoint agents can only be shared within Teams rather than published to Agent Store [10]. Anyone revising for AB-730 should keep the two tools distinct: Agent Builder lives in Copilot Chat and Teams, SharePoint agents live in SharePoint.

When a candidate selects "New agent" in Copilot, three build paths are available [10]:

1. Natural language (recommended): describe the agent conversationally and Copilot fills in the name, description, and instructions automatically, refining them as you add detail. Mentioning email in your description auto-adds "My emails" as a knowledge source, and describing a task like "analyse this Excel data and generate a chart" auto-enables the code interpreter capability.
2. Manual configuration: skip straight to the Configure tab and set every field, including name, icon, default response mode, description, instructions, knowledge, capabilities, and starter prompts, by hand.
3. From a template: Agent Builder ships with templates preconfigured with a description, instructions, and starter prompts for common use cases, which can be used as-is or customised.

Configuring an agent's settings, instructions and capabilities

The third and fourth exam sub-skills cover configuring an agent's knowledge and its settings, meaning instructions, capabilities, and suggested prompts [1]. Every field on the Configure tab carries a specific limit that AB-730 candidates should know precisely rather than approximately [10]:

Instructions direct the agent's behaviour and tasks, and if an agent is built through the natural-language flow, these are auto-generated but remain fully editable afterwards [10]. Starter prompts appear on the Try It tab and help other users understand what the agent is commonly used for, invoking the prompt and starting a conversation the moment a user selects one [10]. Writing effective instructions draws on the same prompt engineering discipline AB-730 tests elsewhere in the syllabus; see Examinotion's Microsoft 365 Copilot prompt engineering guide for techniques that transfer directly to agent instructions.

The Configure tab also sets a default response mode, a detail candidates sometimes overlook because it sounds cosmetic but is directly testable as an agent "setting". Three modes are available: Auto, which balances speed and depth automatically; Quick response, tuned for fast, concise answers to straightforward questions; and Think deeper, which takes more time for complex, multi-step analysis [10]. A user can override the default at chat time, so the setting only controls what happens by default.

Governance-wise, agents built with Agent Builder are covered by a standard Microsoft 365 Copilot licence and do not consume the tenant's Dataverse storage entitlement [10]. One limitation worth remembering for the exam: agents created in Agent Builder cannot currently be used inside Teams Chat [10].

Adding knowledge sources

Configuring an agent's knowledge is one of the most precisely testable parts of this objective, because Microsoft documents exact numeric limits per source type. The current, most granular figures, verified against Microsoft Learn on 4 September 2026, are as follows [11]:

A genuine inconsistency exists across Microsoft's own documentation here, and it is worth understanding rather than memorising a single wrong number. One Agent Builder page still states a flat "up to 20 knowledge sources" limit covering SharePoint sites, folders, and files [10], while the dedicated knowledge-sources article, updated more recently and with far more granularity, gives the per-type figures in the table above [11]. The "20" figure most likely describes the embedded/uploaded-file limit rather than SharePoint or OneDrive specifically. For exam purposes, and for real-world administration, the per-type numbers are the ones that actually govern what an agent can ingest.

A few source-specific rules are worth flagging individually. Public URLs must be no more than two levels deep and cannot contain query parameters [11]. Selecting a SharePoint site does not automatically include its lists, which must be added separately by URL [11]. Email cannot be scoped at all: once added as a knowledge source, the agent uses the entirety of the user's mailbox, and this requires a Copilot add-on licence [11]. SharePoint and OneDrive knowledge sources both require the signed-in user to hold an active Copilot licence themselves; without one, requests fail silently with a generic error [12]. Newly added SharePoint or OneDrive files can take several minutes to become usable, appearing as "Preparing" in the meantime [11].

Sharing an agent with team members

The fifth and final testable skill is sharing an agent [1]. Agent Builder uses a two-role permission model for individual and group sharing [13]:

• Can edit (Owner): full rights to chat with, edit, share, and manage the agent. Only individual users, never groups, can hold this role.
• Can chat: can interact with the agent but cannot edit or manage it. Groups added to an agent can only ever be assigned this role.

An agent starts private, visible only to its creator, until the creator opens the Share dialog, adds people by name, group, or email, assigns a role, and optionally sends an email notification [13]. Owners can change anyone's role, or remove access entirely, and changes take effect immediately [13]. The Share dialog also generates two direct links, a chat link and an edit link, for quick distribution [13].

Making an agent available to the whole organisation uses a separate toggle inside the same Share dialog. Turning on org-wide sharing makes the agent discoverable in Agent Store for everyone in the tenant, but this is distinct from submitting the agent to the organisation's catalogue: the shared version and the catalogue version are managed independently, and only a catalogue submission requires admin review before the agent appears in the "Built by your org" section of Agent Store [13].

Tenant administrators control who is even allowed to share agents in the first place, through a three-option governance model in the Copilot Control System [13]:

• All users (the default): everyone can share agents with the organisation.
• Specific users or groups: sharing rights limited to selected people.
• No users: org-wide sharing disabled completely.

One nuance worth remembering: an agent's knowledge access always respects the end user's own permissions, not the sharer's. If a colleague you share an agent with does not have access to a particular SharePoint file the agent draws on, the agent simply will not surface content from that file for them [13]. Agent Builder can auto-share the specific SharePoint files and folders it was configured to use when you add someone as a chat user or owner, but this auto-sharing behaviour only works when sharing with specific security groups, not when sharing with the whole organisation [13].

Researcher and Analyst: the 2026 reasoning agents

AB-730's audience profile was updated to name two specific agents directly: "experience using generative AI-powered productivity tools, including Microsoft 365 Copilot and agents such as Researcher, and Analyst" [1]. Both are first-party Microsoft experiences that sit alongside standard Copilot chat rather than replacing it, and both are preinstalled and pre-pinned for every Copilot-licensed user through the Microsoft-installed agent mechanism described above [3].

Researcher handles complex, multi-step research tasks, drawing on both the web and a user's own work content, including files, emails, meetings, and chats they already have access to, and returning a source-cited report [14]. Microsoft's own guidance frames the choice clearly: use standard Copilot chat for a quick summary or lightweight brainstorming, and use Researcher when the task needs deeper reasoning across multiple sources or a citation-backed report worth sharing [14]. Usage is capped at 25 queries per user per month, and response times typically range from under five minutes for simple queries up to 45 minutes for genuinely complex ones [15]. Researcher is found under Agents in Copilot Chat, on both desktop and the mobile app [14].

Analyst is positioned as the data-analysis counterpart, better suited than Researcher for spreadsheet-related tasks [15]. It can calculate statistics, identify trends, and surface outliers across Excel spreadsheets, CSV files, and other data sources, producing reports with charts and tables [16]. At the time of writing, Microsoft has not published a dedicated Learn overview page for Analyst, so its exact underlying model and general-availability date are not stated precisely in any Tier 1 source and should not be treated as confirmed facts; readers who need Analyst's practical usage steps should treat it as a reasoning agent focused on data work rather than research synthesis.

Admin control over Researcher and Analyst has a genuine subtlety worth knowing for the exam. One Microsoft Learn page states plainly that both agents "will not fall under any agent-related settings" [4], while the Researcher FAQ states that "admins can block Researcher and Analyst agents" [15]. Read together, the fair synthesis is that Researcher and Analyst sit outside the standard per-agent assignment and deployment controls that apply to other agents, but administrators retain a dedicated, separate ability to block them tenant-wide. Individual users cannot disable or unpin either agent themselves [15].

Frequently Asked Questions

What is the difference between Agent Store and Agent Builder?

Agent Store is where users discover and install agents that already exist, built by Microsoft, partners, or their own organisation. Agent Builder is the tool used to create a new, custom declarative agent from scratch, from a template, or by describing it in natural language, typically for a specific team or task.

How many knowledge sources can a Microsoft 365 Copilot agent have?

The limit depends on the source type: up to four public website URLs, 100 SharePoint files, one SharePoint list, 50 OneDrive files, five Teams chat links, and 20 embedded or uploaded files per agent. There is no single flat number that applies across every source type.

Can I share an agent with a whole team using a group?

Yes, but a group can only be assigned the "Can chat" role, not "Can edit". Only individual users can be made owners of an agent, so if you need someone to manage or edit a shared agent, add them individually rather than through a group.

Is the Project Manager agent still called that in Microsoft 365 Copilot?

Microsoft's current documentation calls this agent "Planner Agent", not "Project Manager agent". The older name still appears in some marketing material and third-party sources, so AB-730 candidates should recognise both names refer to the same agent.

Do Researcher and Analyst count against an agent installation limit?

No. Researcher and Analyst are installed through a separate Microsoft-installed mechanism and are preinstalled and pre-pinned for every Copilot-licensed user rather than being added manually like other Agent Store agents. Admins can block them tenant-wide, but individual users cannot disable or unpin them.

What happens if I share a knowledge source I don't have full access to?

Nothing changes for the person you share the agent with. Agent knowledge always respects the individual end user's own permissions, so if a colleague lacks access to a specific file the agent references, the agent simply omits that content from its responses to them.

Preparing for AB-730's Copilot agents objective

Agent Store, Agent Builder, knowledge configuration, and sharing sit inside a domain worth 35-40% of the AB-730 exam, which makes this one topic worth genuine hands-on practice rather than passive reading. Try building a simple agent from a template in your own tenant if you have access, add a knowledge source or two, and walk through the sharing flow so the roles and limits become familiar rather than memorised. If you want a broader introduction before drilling into agents specifically, Examinotion's Microsoft Copilot agents business guide covers the wider agent landscape beyond just the exam scope, and the AB-730 common exam mistakes article flags where candidates typically lose marks on this domain.

Examinotion's AB-730 practice questions are built around exactly this kind of scenario-based testing, and the full guide to passing AB-730 walks through every domain in the skills outline. Start practising for AB-730 to work through realistic Copilot agent questions alongside the rest of the AI Business Professional skills outline, or browse Examinotion's full exam catalogue if you are still deciding which Microsoft AI certification to prepare for first.

Sources

1. Study guide for Exam AB-730: AI Business Professional — Microsoft Learn, accessed 2026-09-04
2. Agent Store in Microsoft Copilot — Microsoft Learn, accessed 2026-09-04
3. Agent installation in Microsoft Copilot — Microsoft Learn, accessed 2026-09-04
4. Agents admin guide for Microsoft 365 — Microsoft Learn, accessed 2026-09-04
5. Manage agents in the Microsoft 365 admin center — Microsoft Learn, accessed 2026-09-04
6. Set up Facilitator in Microsoft Teams — Microsoft Learn, accessed 2026-09-04
7. Turn off or restrict access to Planner Agent — Microsoft Learn, accessed 2026-09-04
8. Manage Interpreter agent for your organization — Microsoft Learn, accessed 2026-09-04
9. Agent Builder in Microsoft 365 Copilot — Microsoft Learn, accessed 2026-09-04
10. Build Agents with Agent Builder in Microsoft 365 Copilot — Microsoft Learn, accessed 2026-09-04
11. Add knowledge sources to your declarative agent in Microsoft 365 Copilot — Microsoft Learn, accessed 2026-09-04
12. Add knowledge sources to your declarative agent — Microsoft Learn, accessed 2026-09-04
13. Share and manage agents built with Microsoft 365 Copilot — Microsoft Learn, accessed 2026-09-04
14. What is Researcher Agent in Microsoft Copilot? — Microsoft Learn, accessed 2026-09-04
15. Microsoft Copilot Researcher agent frequently asked questions — Microsoft Learn, accessed 2026-09-04
16. Get started with Analyst in Microsoft 365 Copilot — Microsoft Support, accessed 2026-09-04