2.2 Evaluating Built-in Agents for Business Solutions

AB-410 expects you to evaluate the agents Microsoft already ships before recommending that anyone build a custom one. The exam's recurring pattern is a business need plus four agent options, and the correct answer is usually the cheapest option that fully satisfies the requirement — over-building is a real distractor.

Copilots Already Inside the Platform

Before you build anything, inventory the Copilot assistants embedded in Power Platform:

• Copilot in Power Apps (maker) — describe the app you want at make.powerapps.com and Copilot drafts Dataverse tables, sample data, and screens; it also edits formulas and answers how-to questions while you build.
• Copilot control for app users — drop the control into a canvas app and end users chat over the app's Dataverse data: summarize this record, filter these rows, draft an update.
• Copilot in model-driven apps — a sidecar chat that answers questions about the records in view and suggests next steps.
• Copilot in Power Automate — describe the automation in natural language and Copilot assembles the cloud flow, suggests actions, and fills parameters.
• Copilot in the Power Pages design studio — generates sitemaps, pages, copy, forms, and a starter chatbot for the site.
• Dataverse AI experiences — create tables from an Excel upload or a sentence, and generate row summaries.

These assistants cost no design time. The evaluation question is never whether a Copilot can help — it is whether an out-of-box agent fully satisfies the requirement.

Custom Agents in Copilot Studio

When the answer is no, the next rung is Microsoft Copilot Studio at copilotstudio.microsoft.com, the low-code studio for custom agents. Evaluate its three capability pillars.

Knowledge sources

Ground generative answers in SharePoint and OneDrive, public websites, uploaded documents, Dataverse, and Graph connectors, so the agent answers from your content instead of the open model.

Actions

Let the agent do work, not just talk: call any of the 1,000+ connectors, run Power Automate cloud flows, execute AI Builder prompts, or call REST APIs. A requirement that says the assistant must create, update, or submit anything implies actions — which immediately rules out knowledge-only options like SharePoint agents.

Channels

Publish one agent to Microsoft Teams and Microsoft 365 Copilot, a standalone web app, a custom website, or a Power Pages site.

Generative orchestration vs classic topics

Agents mix classic topics (authored dialog trees fired by trigger phrases) with generative orchestration, where a planner reasons over instructions, knowledge, topics, and actions to satisfy the request. Generative orchestration is the default for new agents and is what makes them feel agentic rather than scripted.

Autonomous Agents and Triggers

Autonomous agents remove the human from the start of the conversation. Instead of waiting for a chat message, the agent starts from an event trigger — a new email in a shared mailbox, a new row in Dataverse, a file landing in SharePoint, or a recurrence — then plans and executes its steps, escalating to a human when its instructions say so. You add triggers from the agent's Overview page in Copilot Studio and watch each run in the activity view. Exam signal: any requirement phrased as 'when X happens, the system should research, decide, and act without being asked' points to an autonomous agent with event triggers, not a chatbot with topics.

Declarative Agents for Microsoft 365 Copilot

Declarative agents extend Microsoft 365 Copilot itself. They are declared with instructions, knowledge (SharePoint, OneDrive, Graph connectors, web content), starter prompts, and optional actions, and they run on the same orchestrator as Microsoft 365 Copilot — inheriting its security, compliance, and grounding in Microsoft Graph. The maker-friendly path is agent builder, embedded in the Microsoft 365 Copilot experience: describe the agent, point it at content, publish, no code. Copilot Studio and the pro-code Microsoft 365 Agents Toolkit produce the same agent type with more control. Use a declarative agent when users already live in Microsoft 365 Copilot or Teams chat and the job is grounded Q&A plus light actions. Exam limitation: declarative agents live only inside Microsoft 365 Copilot experiences — they are not standalone web chat bots.

Agents in SharePoint

SharePoint agents are the fastest built-in option: every SharePoint site can have an agent grounded in that site's content, and users can scope an agent to selected libraries, folders, or files. They respect SharePoint permissions, so answers are trimmed to what the asker is allowed to see, and they are shared like any other SharePoint asset. Evaluate them for 'answer questions about this site's documents' and nothing more — a SharePoint agent cannot reach Dataverse, cannot call actions, and cannot be published to custom channels.

The Evaluation Ladder

Cost belongs in the evaluation. Copilot Studio consumption is metered in Copilot Credits, sold in tenant capacity packs of 25,000 credits per month (with pay-as-you-go through an Azure subscription as the alternative), while embedded Copilots ride along with product licenses. Walk the ladder from the bottom and stop at the first rung that satisfies every stated requirement.

Exam Traps

• Recommending a custom Copilot Studio agent when the built-in Copilot in a canvas or model-driven app already answers the question.
• Forgetting that SharePoint agents are Q&A-only and permission-trimmed — they cannot act on Dataverse or other systems.
• Placing a declarative agent on a standalone website; declarative agents exist only inside Microsoft 365 Copilot.
• Confusing autonomous event triggers with classic trigger phrases: triggers start runs without any chat message.
• Assuming each channel needs a separate agent build — one Copilot Studio agent publishes to many channels.