2.4 Developing Use Cases for Prebuilt Agents & the Microsoft Agent Catalog

Prebuilt Agent Use Cases: The Microsoft Agent Catalog & Fit Assessment

Quick Answer: Before an architect authorises a single line of custom agent configuration, the AB-100 blueprint expects a deliberate survey of what Microsoft already ships. Prebuilt agents arrive in four distinct supply channels: first-party Dynamics 365 agents that are switched on inside a business app (Sales Qualification Agent, Supplier Communications Agent, Customer Intent), Microsoft 365 Copilot agents licensed with the Copilot seat (Researcher, Analyst, Facilitator, SharePoint agents), Copilot Studio agent templates that ship as editable starting points, and Dynamics 365 agent templates published through the FastTrack implementation-assets repository on GitHub for financial services, retail, healthcare, and the public sector. The architectural discipline is not "which agent is coolest" but a structured fit assessment: does a shipped agent already cover the business outcome at an acceptable configuration cost, and does adopting it transfer the responsible-AI, ALM, and update burden to Microsoft rather than to the customer?

Most failed agentic programmes do not fail because the technology did not work. They fail because a delivery team spent an eleven-week sprint hand-building a conversational lead-qualification agent in Copilot Studio that Microsoft already shipped, supported, and updated quarterly inside Dynamics 365 Sales. The custom build then decayed: nobody owned its prompts, its knowledge index drifted, and its behaviour diverged from the platform agent that colleagues in another business unit had simply enabled.

Developing the use cases for prebuilt agents is therefore a planning-domain activity, not a design-domain one. It happens before the solution architecture is drawn, and its output is a written inventory that says, for each candidate business process, whether the organisation will enable, template, extend, or build.

---

1. The Four Supply Channels for Prebuilt Agents

1.1 First-Party Dynamics 365 Agents

These are Microsoft-authored agents that live inside a Dynamics 365 application and are enabled through that application's admin centre. They are not downloaded, not imported, and not versioned by the customer. Microsoft owns the model selection, the orchestration, the responsible-AI review, and the release cadence.

Representative examples an AB-100 candidate should recognise by business outcome:

Release status varies by agent — several are generally available while others, such as the Account Reconciliation Agent and the Expense Agent, are in preview at the time of writing, and Microsoft notes that most exam questions cover generally available features while preview features may appear if they are commonly used. The catalogue expands every release wave. An architect should never memorise it as a fixed list; the exam-relevant skill is knowing where to look (the app's own admin centre plus the current Dynamics 365 release plan) and knowing that first-party agents are enabled, not deployed.

1.2 Microsoft 365 Copilot Agents

These ship with the Microsoft 365 Copilot licence and operate across Microsoft Graph rather than inside a line-of-business application:

• Researcher — performs multi-step research across work data, the web, and connected sources, producing a cited synthesis. Use it when the requirement is planning, multi-source retrieval, and reasoning, not a single lookup.
• Analyst — turns tabular business data into insight and visualisations, iterating over analysis steps the way a data analyst would.
• Facilitator — takes notes, answers in-meeting questions, and moderates Microsoft Teams meetings.
• SharePoint agents — scoped to a site, library, or set of documents, answering questions grounded strictly in that content with the user's existing permissions honoured.

The critical architectural point: these agents are already grounded in tenant content through Microsoft Graph and already permission-trimmed. Rebuilding "an agent that answers questions about our SharePoint policies" as a custom Copilot Studio agent frequently reproduces, badly, what a SharePoint agent does natively.

1.3 Copilot Studio Agent Templates

Copilot Studio publishes editable templates that pre-populate topics, instructions, tools, and sample knowledge configuration for common scenarios such as IT helpdesk, HR benefits, safety reporting, and website support. Unlike first-party Dynamics agents, a template produces a customer-owned agent: the moment it is created, the organisation owns its lifecycle, its solution packaging, and its responsible-AI review. A template shortens time-to-first-demo; it does not reduce long-run ownership cost.

1.4 Dynamics 365 Agent Templates (FastTrack)

Microsoft publishes industry agent templates for financial services, retail, healthcare, and the public sector through the Dynamics 365 FastTrack implementation-assets repository on GitHub, with product-specific families for Service, Commerce, and Customer Insights. Microsoft describes their benefits as a faster start without complex architecture choices, industry-focused scenarios that address frequently requested regulatory requirements, deployment in days, a known high-impact process with identifiable ROI, and standardised behaviour and governance patterns across teams.

Because these are downloaded solution artefacts, they behave like templates for ALM purposes: the customer imports, customises, and thereafter owns them.

---

2. The Fit Assessment: A Four-Gate Decision

A defensible prebuilt-agent recommendation runs each candidate business process through four sequential gates. Failing any gate pushes the process to the next tier.

Gate 1 — Outcome coverage. Does a shipped agent already deliver the business outcome, not merely a superficially similar feature? Compare against the process's exit criteria, not its narrative. "Chases suppliers for delivery confirmation" is an outcome; "talks to suppliers" is not.

Gate 2 — Data and licensing preconditions. First-party agents assume their host application's data model. The Supplier Communications Agent assumes purchase orders in Dynamics 365 Supply Chain Management. If purchase orders live in SAP and are only mirrored into Dataverse as virtual tables, the prebuilt agent may not see the fields it needs. Equally, Researcher and Analyst require Microsoft 365 Copilot seats for the users who will invoke them.

Gate 3 — Configuration headroom. Every first-party agent exposes a bounded configuration surface — instructions, prompts, thresholds, knowledge attachment, approval requirements. If the business requirement falls inside that surface, enabling is correct. If it requires behaviour the surface cannot express, the process escalates to extend or build.

Gate 4 — Governance transfer. Enabling a first-party agent transfers responsible-AI review, model currency, content-filter tuning, and regression testing to Microsoft. Building transfers all of it to the customer. For a regulated process, this gate frequently dominates the other three even when a custom build would be marginally more capable.

---

3. Worked Fit Assessment: A Distribution Business

A European industrial distributor brings four processes to its AI Center of Excellence intake board.

Three of four processes are delivered by enabling shipped agents. The single custom build absorbs the whole engineering budget — and, crucially, the architect can now justify that budget because the alternative was documented and rejected on evidence.

---

4. Anti-Patterns the Exam Tests

• Catalogue blindness. Designing a custom Copilot Studio agent for lead qualification, case triage, or document Q&A without first checking whether Dynamics 365 or Microsoft 365 Copilot already ships it. This is the single most common wrong answer stem in planning-domain scenarios.
• Template mistaken for first-party. Assuming that because an agent came from a Microsoft template, Microsoft will keep it current. A template creates a customer-owned artefact from the moment of instantiation.
• Ignoring the licence precondition. Proposing Researcher or Analyst for users without Microsoft 365 Copilot seats, or proposing Dynamics-hosted agents to users without the relevant Dynamics 365 licence.
• Enabling everything at once. Turning on the full first-party catalogue across production without a pilot cohort, feedback loop, or measurement plan produces adoption noise and no attributable value.
• Treating "prebuilt" as "no governance". Even enabled first-party agents need an owner, an adoption metric, an escalation path, and a review of what grounding data they now reach.

AB-100 Exam Tip: When a planning scenario describes a common, well-bounded business process inside a Dynamics 365 app — lead qualification, case summarisation, supplier follow-up, ledger reconciliation — and the option set offers both "build a custom agent in Copilot Studio" and "enable the corresponding first-party agent and configure it", the shipped agent is almost always the intended answer. Custom build becomes correct only when the scenario explicitly introduces logic, data, or a system that lies outside the host application.