Copilot Studio Companion

Evaluate scenarios, explore capabilities, navigate governance, and design agent architectures

55 scenarios evaluated ☕ Support
I am:

What do you need?

Toggle the features your agent needs — each platform lights up green (supported), amber (partial), or red (not available).

Agent Builder
Copilot Studio
AI Foundry
All platforms can do this — Agent Builder is the easiest starting point
Agent Builder and CS have native M365 integration
Agent Builder has no topics or conversation flow control
CS has 1,000+ connectors; AI Foundry has full API access
Only Copilot Studio has native Power Automate integration
Agent Builder is M365 only; CS and Foundry support multi-channel
Agent Builder uses the default model only
Autonomous agents run without a user starting the conversation
CS supports child, connected, and cross-platform agents
CS and Foundry have enterprise governance; Agent Builder is limited
Only AI Foundry supports custom model training
Adaptive Cards let agents show forms, buttons, and structured data
Copilot Studio can operate desktop apps like a human user
Only AI Foundry supports sovereign/air-gapped environments
AI Foundry requires developer skills
Agent Builder is free with M365 Copilot; CS needs its own licence

Toggle features above to see which platform fits your needs.

🔍 Describe Your Scenario

Common Scenarios

Browse pre-built scenarios to see which platform is right — and which components you'll need.

👥

HR FAQ Bot

Start with Agent Builder

Employees ask about leave policies, benefits, onboarding — agent answers from HR SharePoint.

Why this platform?

This is simple Q&A from company documents. Agent Builder handles this perfectly — no need for Copilot Studio's complexity.

Key components:
  • Knowledge sources (SharePoint)
  • Instructions

⬆️ When you need approval workflows (e.g. leave requests) → upgrade to Copilot Studio

🎧

IT Helpdesk Agent

Copilot Studio

Reset passwords, create tickets in ServiceNow, escalate to humans, track resolution status.

⚠️ Premium connector required (ServiceNow)⚠️ Credit-intensive (GenAI + tools per interaction)
Why this platform?

This needs external system integration (ServiceNow), multi-step workflows (ticket creation → assignment → status check), and Power Automate flows. Classic Copilot Studio territory.

Key components:
  • Topics
  • Power Automate flows
  • ServiceNow connector
  • Adaptive Cards
  • Human handoff
💬

External Customer Support Bot

Copilot Studio

Customer-facing bot on your website answering product questions, handling returns, collecting feedback.

⚠️ External data boundary⚠️ Unpredictable credit consumption⚠️ Web channel security required
Why this platform?

External-facing agents need multi-channel deployment (website widget), unauthenticated access, and likely custom connectors to your CRM/order system.

Key components:
  • Web channel
  • Knowledge sources
  • Custom connector (CRM)
  • Unauthenticated access
  • Adaptive Cards
✅

Multi-Step Approval Workflow

Copilot Studio

Purchase requests that need manager approval, budget check, then procurement — all through chat.

Why this platform?

Approval workflows require Power Automate integration, Dataverse for tracking, and multi-step conversation topics with conditional branching.

Key components:
  • Topics with branching
  • Power Automate (Approvals)
  • Dataverse
  • Adaptive Cards (forms)
📊

CRM Integration Agent

Copilot Studio

Sales team asks the agent about customer history, pipeline, upcoming renewals — pulls live data from Dynamics 365 or Salesforce.

⚠️ Premium connector required⚠️ Admin consent for CRM access
Why this platform?

Live CRM data integration requires connectors (Dynamics 365 or custom Salesforce connector) and structured conversation flows to present data meaningfully.

Key components:
  • Connectors (Dynamics 365 / Salesforce)
  • Adaptive Cards
  • Topics
  • Generative orchestration
🎓

Employee Onboarding Guide

Copilot Studio

New hires get guided through setup — IT equipment, badge, training schedule, benefits enrolment.

Why this platform?

Onboarding needs multi-step guidance with conditional paths (different by role/department), task tracking via Dataverse, and potentially Power Automate for ticket creation.

Key components:
  • Topics (role-based branching)
  • Power Automate
  • Dataverse
  • Knowledge sources
📝

Meeting Notes Summariser

Start with Agent Builder

Agent that summarises meeting transcripts and extracts action items from Teams meetings.

Why this platform?

Agent Builder can access Teams meeting transcripts via M365 Graph Grounding. Simple instruction-based agent — no workflows needed.

Key components:
  • Instructions
  • M365 Graph Grounding

⬆️ When you need to auto-create tasks in Planner or send follow-up emails → upgrade to Copilot Studio

🌐

Teams + Website + WhatsApp Bot

Copilot Studio

Same agent deployed across Teams for employees, website for customers, and WhatsApp for mobile.

Why this platform?

Multi-channel deployment is a Copilot Studio feature. Agent Builder agents only live inside M365 apps.

Key components:
  • Multi-channel publishing
  • Teams channel
  • Web widget
  • WhatsApp channel
  • Authentication config per channel
📈

Custom AI Data Analysis Pipeline

Azure AI Foundry

Process thousands of documents, extract structured data, run custom ML models, generate reports.

Why this platform?

This needs custom model training, RAG pipelines at scale, and programmatic orchestration. Beyond what Copilot Studio's low-code canvas can handle.

Key components:
  • Custom models
  • RAG pipeline
  • Azure AI Search
  • Custom evaluation
  • CI/CD
📋

Company Policy Lookup

Start with Agent Builder

Employees ask about travel policy, expense rules, compliance requirements — agent answers from internal docs.

Why this platform?

Pure knowledge retrieval from company documents. Agent Builder with SharePoint knowledge sources handles this perfectly.

Key components:
  • Knowledge sources (SharePoint)
  • Instructions

⬆️ When you need to collect expense reports or route approvals → upgrade to Copilot Studio

🤖

Autonomous Monitoring Agent

Copilot Studio

Agent monitors Dataverse for overdue tickets, sends reminders, escalates to managers — all without user prompts.

Why this platform?

Autonomous agents with event triggers are a Copilot Studio capability. The agent runs in the background, monitoring and acting on data changes.

Key components:
  • Autonomous triggers
  • Dataverse
  • Power Automate
  • Adaptive Cards
🏛️

Government / Sovereign AI Deployment

Azure AI Foundry

AI agent for a government agency requiring air-gapped deployment, data sovereignty, and specific compliance certifications.

Why this platform?

Sovereign and air-gapped deployments require Azure AI Foundry. Copilot Studio doesn't support isolated environments.

Key components:
  • Azure AI Foundry
  • Sovereign cloud
  • Custom models
  • Compliance certifications

💰 Quick Cost Comparison

Included Included with M365 Copilot ($30/user/mo)

No extra cost if you already have M365 Copilot licences.

Credit-based $200/mo for 25K credits (or $0.01/credit PAYG)

Regular messages = 1 credit. GenAI answers = 2 credits. Complex interactions with tools can use 20+ credits.

Consumption Per-token pricing (varies by model)

Pay per token, per tool invocation, per storage unit. Full Azure consumption model.

🚀 Open Copilot Studio 🤖 Agent Builder Guide — the simpler option 📖 Read the full comparison blog

Copilot Studio Capabilities

Everything Copilot Studio can do, explained for IT admins — not developers.

📊 Quick Comparison: Agent Builder vs Studio vs Foundry
CapabilityAgent BuilderCopilot StudioAI Foundry
Topics & conversation flow❌✅✅ (code)
Knowledge grounding✅ M365✅ Multi-source✅ Custom RAG
Model selection❌ Default only✅ 6+ models✅ 1800+ models
Adaptive Cards❌✅✅ (custom)
Connectors (1000+)❌✅Via code
Power Automate❌✅ Native❌
Tools & MCP servers❌✅✅
Multi-channel deploy❌ M365 only✅ 8+ channels✅ Any
Computer Use❌✅❌
Fine-tuning❌❌✅
No-code authoring✅✅❌
💬 Conversation Topics 📚 Knowledge Sources 🧠 AI Model Selection 🃏 Adaptive Cards 🔌 Connectors & Power Automate 🛠️ Tools & MCP Servers 📡 Deployment Channels

💬 Conversation Topics

💡 Think of topics like a phone menu system — but smarter. Instead of 'Press 1 for billing', the AI understands what the user wants and routes them to the right conversation path automatically.

Topics are structured conversation flows with trigger phrases, branching logic, conditions, variables, and actions. They're the building blocks of how your agent responds to specific intents.

When to use this

Use authored topics when you need precise control over a conversation — like guiding someone through a ticket submission, an approval, or a multi-step process. Use generative orchestration to let AI pick the right topic automatically.

Key Facts

  • Two types: authored (you build them) and system (pre-built, like Greeting and Fallback)
  • Generative orchestration lets AI automatically choose which topic to use — no manual routing
  • Topics can call Power Automate flows, show Adaptive Cards, set variables, and branch on conditions
  • You can test topics in real-time without publishing the agent
  • Trigger phrases tell the AI when this topic should activate — add 5-10 varied phrases
⚠️ Gotchas & Limitations
  • Too many topics with overlapping triggers causes confusion — keep triggers distinct
  • System topics (Greeting, Escalation, Fallback) should be customised, not left as defaults
  • Topic names appear in analytics — use clear, descriptive names from the start
📄 Microsoft Learn docs →

📚 Knowledge Sources

💡 Knowledge sources are like giving your agent a library card. You tell it which bookshelves to read from (SharePoint sites, websites, Dataverse), and it finds answers from those sources using AI.

Knowledge sources ground your agent's responses in real data. The agent reads your documents and generates natural language answers with citations. This is 'generative answers' — the agent doesn't just match keywords, it understands and summarises.

When to use this

Use knowledge sources for any Q&A scenario — HR policies, IT documentation, product manuals, compliance guides. The agent reads your docs so users don't have to.

Key Facts

  • Supported sources: SharePoint sites, SharePoint lists, Dataverse tables, public websites, uploaded files
  • With M365 Copilot licence: files up to 200MB supported via Tenant Graph Grounding
  • Without M365 Copilot: file size limit is 7MB per document
  • Sync frequency: SharePoint sources refresh every 4-6 hours
  • Max 1,000 files per SharePoint site, 50 folders, 10 levels of subfolder depth
  • Supported file types: DOC, DOCX, PPT, PPTX, PDF, XLS, XLSX
⚠️ Gotchas & Limitations
  • Only modern SharePoint pages — classic ASPX pages are ignored
  • Sensitivity labels (Confidential/Highly Confidential) block indexing
  • Password-protected files show as 'Ready' but return no answers
  • Ampersands (&) in document or folder names are not supported
  • Guest users cannot use generative answers from SharePoint in SSO-enabled apps
📄 Microsoft Learn docs →

🧠 AI Model Selection

💡 Choosing a model is like picking a worker for a job. Mini models are like interns — fast and cheap for simple tasks. General models are experienced staff. Deep models are senior consultants for complex analysis.

Copilot Studio lets you choose which AI model powers your agent. Different models offer different trade-offs between speed, quality, and cost (credit consumption).

When to use this

Use Mini (GPT-4.1 mini) for simple FAQ bots. General (GPT-5, Claude) for content creation. Deep (o3, Claude Opus) for complex reasoning. Or set Auto and let the platform decide per query.

Key Facts

  • Four model tiers: Mini (fast/cheap), General (balanced), Deep (reasoning), Auto (adaptive per turn)
  • Models available: GPT-4.1, GPT-4.1 mini, GPT-5, Claude Sonnet 4.5, Claude Opus 4, Grok 3
  • Default is GPT-4.1 mini — change this if your agent needs higher quality
  • Auto mode picks the best model per turn — fast models for simple questions, powerful models for complex ones
  • Model selection affects credit consumption — Deep models cost more credits per response
⚠️ Gotchas & Limitations
  • Not all models are available in all regions — check your environment's geographic region
  • Some models require cross-geo data movement approval from your admin
  • Claude models enable Computer Use capability — consider this for desktop automation scenarios
  • Changing models mid-deployment may alter response quality — test thoroughly before switching
📄 Microsoft Learn docs →

🃏 Adaptive Cards

💡 Adaptive Cards turn chat messages into mini web pages. Instead of plain text saying 'Here are 3 options', the agent shows a visual card with buttons, forms, images, and choices — like a small app inside the chat.

Adaptive Cards are JSON-based UI elements that render rich, interactive content within Teams, websites, and other channels. They can display data, collect input, and trigger actions.

When to use this

Use Adaptive Cards when plain text isn't enough — showing product details with images, collecting form data (expense reports, survey responses), presenting comparison tables, or giving users structured choices.

Key Facts

  • Supported in Teams, web widget, and most Direct Line channels
  • Can collect user input: text fields, dropdowns, date pickers, toggle switches
  • The generative builder can help create Adaptive Cards from natural language descriptions
  • Cards can include action buttons that trigger Power Automate flows or topic transitions
  • Maximum card payload: 40KB (per the Bot Framework limit)
⚠️ Gotchas & Limitations
  • Card rendering varies by channel — test in every channel you plan to deploy to
  • Not all channels support all card features (e.g. WhatsApp has limited card support)
  • Complex cards with many inputs can feel overwhelming — keep them focused on one task
  • Cards are stateless — data must be captured via variables when submitted
📄 Microsoft Learn docs →

🔌 Connectors & Power Automate

💡 Connectors are like USB adapters for your agent. Each connector lets your agent plug into a different system — ServiceNow, Salesforce, SQL databases, or any REST API. Power Automate is the extension cord that chains these connections into workflows.

Copilot Studio integrates with 1,000+ Power Platform connectors (Standard, Premium, Custom) and can trigger Power Automate cloud flows as actions. This lets agents read from and write to business systems.

When to use this

Use connectors when your agent needs to interact with external systems. Use Power Automate when you need multi-step workflows — create a ticket, send an email, update a record, run an approval.

Key Facts

  • Standard connectors: SharePoint, Outlook, Teams, OneDrive, Excel, Planner
  • Premium connectors: Dataverse, SQL Server, Dynamics 365, ServiceNow, Salesforce
  • Custom connectors: wrap any REST API as a reusable connector
  • Power Automate flows run within the Copilot Studio context — no separate licence needed for cloud flows triggered by agents
  • HTTP Request node lets you call any REST endpoint directly without a connector
⚠️ Gotchas & Limitations
  • DLP policies control which connectors agents can use — check with your admin
  • Connector authentication uses the connection owner's credentials unless delegation is configured
  • Premium connectors require appropriate Power Platform licensing in the environment
  • Long-running flows (>2 minutes) may need async patterns — the agent can't wait indefinitely
📄 Microsoft Learn docs →

🛠️ Tools & MCP Servers

💡 Tools are like giving your agent a Swiss Army knife. Each tool is a specific skill — calling an API, running a calculation, looking up data. MCP servers are external toolboxes that can be shared across multiple agents.

Agent tools extend what your agent can do beyond conversation. Tools can call APIs, run Power Automate flows, or connect to external MCP (Model Context Protocol) servers. With generative orchestration, the AI automatically decides which tool to use.

When to use this

Add tools when your agent needs to take actions — calculate shipping costs, check inventory, create records. Use MCP servers when you want to share a set of tools across multiple agents or platforms.

Key Facts

  • Tools can be: Power Automate flows, custom connectors, HTTP actions, or MCP server connections
  • Generative orchestration automatically selects the right tool based on user intent
  • MCP servers expose tools via a standardised protocol — build once, use everywhere
  • Computer Use is a special tool that lets agents operate desktop applications visually
  • Tools can have input parameters and return structured output for the agent to use
⚠️ Gotchas & Limitations
  • Tool descriptions are critical — the AI uses them to decide when to invoke each tool
  • Too many tools can confuse the orchestrator — keep tool sets focused and well-described
  • MCP server connections require network access from the Copilot Studio environment
  • Computer Use requires Claude Sonnet 4.5 model and is best for desktop-only workflows without APIs
📄 Microsoft Learn docs →

📡 Deployment Channels

💡 Channels are like opening branches of a shop. Your agent is the same business, but it serves customers in Teams (the office branch), your website (the high street branch), WhatsApp (the mobile branch), and more.

Copilot Studio agents can be published to multiple channels simultaneously. Each channel may have different authentication, appearance, and capability settings.

When to use this

Deploy to Teams for internal employees. Add a web widget for website visitors. Enable WhatsApp or Facebook for customer support on mobile. Use Dynamics 365 Omnichannel for unified customer service.

Key Facts

  • Available channels: Teams, Web widget, Mobile app, Facebook, WhatsApp, Azure Bot Service, Dynamics 365 Omnichannel, SharePoint, Custom (Direct Line)
  • Teams deployment: publish to your org's app catalogue (admin approval may be required)
  • Web widget: embed with a JavaScript snippet — customise appearance and behaviour
  • Each channel can have different authentication requirements (SSO, manual auth, or none)
  • DLP policies can block specific channels at the environment or tenant level
⚠️ Gotchas & Limitations
  • Not all features work identically across channels — test Adaptive Cards in each channel
  • WhatsApp and Facebook require business verification with Meta
  • Unauthenticated web channels have different security implications — configure web channel security
  • SharePoint channel embeds the agent on SharePoint pages but may require Managed Environments
📄 Microsoft Learn docs →

🔎 Connector Explorer

Search the top 30 connectors. Premium connectors need appropriate Power Platform licensing.

📁 SharePoint Standard
Microsoft 365
  • Document knowledge
  • List data
  • File management
📧 Outlook 365 Standard
Microsoft 365
  • Send emails
  • Calendar lookups
  • Meeting scheduling
💬 Microsoft Teams Standard
Microsoft 365
  • Post messages
  • Channel notifications
  • Adaptive Cards
☁️ OneDrive for Business Standard
Microsoft 365
  • File access
  • Document sharing
  • Knowledge grounding
📋 Planner Standard
Microsoft 365
  • Create tasks
  • Update status
  • Assign work
📊 Excel Online Standard
Microsoft 365
  • Read/write spreadsheets
  • Data lookups
  • Report generation
✅ Approvals Standard
Microsoft 365
  • Request approvals
  • Approval workflows
  • Decision tracking
🗄️ Dataverse Premium
Power Platform
  • Business data storage
  • CRUD operations
  • Relational queries
⚡ Power Automate Standard
Power Platform
  • Trigger flows
  • Complex workflows
  • Scheduled actions
🤖 AI Builder Premium
Power Platform
  • Document processing
  • Object detection
  • Text classification
🛢️ SQL Server Premium
Database
  • Query databases
  • CRUD operations
  • Report data
🛢️ Azure SQL Premium
Database
  • Cloud database queries
  • Stored procedures
  • Data lookups
🌍 Cosmos DB Premium
Database
  • NoSQL queries
  • Document storage
  • Global data access
🎫 ServiceNow Premium
ITSM
  • Create incidents
  • Query tickets
  • Update records
🐛 Jira Premium
ITSM
  • Create issues
  • Sprint tracking
  • Bug management
💼 Salesforce Premium
CRM
  • Account lookups
  • Opportunity management
  • Lead tracking
📈 Dynamics 365 Premium
CRM
  • Customer data
  • Case management
  • Sales pipeline
🏭 SAP Premium
ERP
  • Purchase orders
  • Inventory queries
  • Financial data
🌐 HTTP Premium
Custom
  • Any REST API
  • Webhooks
  • Custom integrations
🔧 Custom Connector Premium
Custom
  • Wrap any REST API
  • Reusable across agents
  • DLP-manageable
🔐 Azure Key Vault Premium
Azure
  • Secret storage
  • Certificate management
  • Secure config
📦 Azure Blob Storage Premium
Azure
  • File storage
  • Blob management
  • Document processing
🧠 Azure OpenAI Premium
Azure
  • Custom prompts
  • Embeddings
  • Model inference
📱 Twilio Premium
Communication
  • SMS notifications
  • Phone calls
  • WhatsApp messages
💬 Slack Premium
Communication
  • Post messages
  • Channel notifications
  • User lookups
✉️ Gmail Standard
Google
  • Send/read emails
  • Label management
  • Search
📊 Google Sheets Standard
Google
  • Read/write rows
  • Data lookups
  • Spreadsheet automation
🎧 Zendesk Premium
ITSM
  • Create tickets
  • Customer lookup
  • Knowledge base
🐙 GitHub Premium
DevOps
  • Create issues
  • PR management
  • Repo queries
🔄 Azure DevOps Premium
DevOps
  • Work items
  • Pipeline triggers
  • Board queries
🚀 Open Copilot Studio 📚 Full documentation

Governance & Sharing

Everything admins need to know about controlling, securing, and sharing Copilot Studio agents.

🚦 Before You Start — Prerequisites

Check these BEFORE building your first agent. Missing prerequisites are the #1 cause of failed Studio projects.

🔑 Copilot Studio Licence Required

Do you have a Copilot Studio tenant licence or pay-as-you-go via Azure?

Details

Without a licence, you can't create agents in Copilot Studio. The tenant licence includes 25K credits/month. PAYG costs $0.01/credit.

📄 Learn more →
🏗️ Power Platform Environment Required

Do you have a non-default Dataverse environment set up for Copilot Studio?

Details

Agents need a Dataverse environment. Don't use the default environment for production agents — create dedicated environments for Dev, Test, and Production.

📄 Learn more →
⚙️ Power Platform Admin Access Required

Does your admin have Power Platform admin or Environment Admin role?

Details

Someone needs admin access to create environments, set DLP policies, manage connectors, and approve agent publishing.

📄 Learn more →
🛡️ DLP Policies Reviewed Required

Have you reviewed which connectors and channels are allowed by DLP?

Details

DLP enforcement is mandatory since early 2025. If your agent needs a connector that DLP blocks, you'll get errors on publish. Review policies BEFORE building.

📄 Learn more →
👥 Entra ID Security Groups Recommended

Do you have security groups set up for agent makers and users?

Details

Use Entra ID groups to control who can create agents, who can access specific environments, and who gets Copilot Studio licences.

📄 Learn more →
🔐 Authentication Plan Recommended

Have you decided how users will authenticate with the agent?

Details

Internal agents: use SSO with Entra ID. External agents: decide between manual auth, unauthenticated access, or partner auth. This affects security and user experience.

📄 Learn more →

🏗️ Power Platform Environments

💡 Environments are like separate offices in a building. Each office has its own security badge (access controls), its own filing cabinets (Dataverse), and its own rules (DLP policies). You build agents in the dev office, test them in the test office, and deploy to the production office.

Every Copilot Studio agent lives inside a Power Platform environment. Environments are containers that hold agents, Dataverse data, connectors, and security settings. Using separate environments for development, testing, and production is a best practice for enterprise deployments.

✅ Admin Checklist

  • Create separate Dev, Test, and Production environments
  • Use Managed Environments for enhanced governance controls
  • Assign Environment Admin and Environment Maker roles to appropriate groups
  • Enable environment routing to direct new makers to designated environments
  • Set up Application Lifecycle Management (ALM) for promoting agents between environments
📄 Microsoft Learn docs →

🛡️ DLP Policies

💡 DLP policies are like customs officers at a border. They check what data is going where and block anything that shouldn't cross the line. They prevent your agent from accidentally sending company secrets to an external API or social media channel.

Data Loss Prevention policies control which connectors and channels agents can use. DLP policies are configured at the tenant or environment level in the Power Platform admin centre. Since early 2025, DLP enforcement is mandatory for all tenants — no exemptions.

✅ Admin Checklist

  • Review which connectors are in Business vs Non-business vs Blocked groups
  • Block channels you don't want agents published to (Facebook, WhatsApp, etc.)
  • Block HTTP connector if agents shouldn't make arbitrary web requests
  • Use endpoint filtering to allow specific HTTP endpoints instead of blocking all HTTP
  • Block event triggers if you want to prevent autonomous agent creation
  • Test DLP policy impact — blocked agents show detailed error messages with policy names
  • Use the Centre of Excellence (CoE) Starter Kit to audit agent-DLP compliance
📄 Microsoft Learn docs →

🤝 Sharing (Internal vs External)

💡 Sharing an agent is like lending a car. Sharing internally (employees) means they drive on company roads with company insurance. Sharing externally (website, WhatsApp) means anyone on the public road can drive — you need different insurance, different rules.

Copilot Studio agents can be shared internally within your organisation or externally with customers and partners. Internal sharing happens via Teams, M365 Copilot, or SharePoint. External sharing uses web widgets, WhatsApp, Facebook, or custom Direct Line channels. Each has different authentication, security, and licensing implications.

✅ Admin Checklist

  • Internal agents: deploy to Teams app catalogue with admin approval
  • Internal agents: can use SSO (Single Sign-On) with Entra ID for seamless auth
  • External agents: configure authentication (Entra ID for partners, unauthenticated for public)
  • External agents: enable web channel security to prevent embedding on unauthorised sites
  • External agents: review and test Adaptive Card rendering in target channels
  • Consider that external-facing agents consume more Copilot Credits (public traffic can spike)
  • Set up monitoring/alerts for unexpected credit consumption from external channels
📄 Microsoft Learn docs →

⚙️ Admin Controls

💡 Admin controls are like a building's master control room. Global Admins and Power Platform Admins can see everything happening with agents across the organisation — who built what, where it's deployed, and what it's accessing.

Administrators manage Copilot Studio through the Power Platform admin centre. They control who can create agents, which environments they work in, what data they can access, and where they can be published. Tenant-level settings override environment-level settings.

✅ Admin Checklist

  • Control who can create agents using Entra ID security groups
  • Enable or disable publishing of generative AI-powered agents at tenant level
  • Manage environment-level GenAI features (can be disabled where regional capacity isn't available)
  • Review agent connections — ensure they use strong authentication and delegation
  • Store sensitive credentials in Azure Key Vault, not in connection configurations
  • Use the Power BI CoE Dashboard for tenant-wide visibility into agents and usage
  • Set up telemetry forwarding to Application Insights for detailed monitoring
📄 Microsoft Learn docs →

🤖 Agent 365 Governance

💡 Agent 365 governance is about knowing where all your agents live. Like a company vehicle fleet — every agent should be registered, insured (licensed), and have a responsible owner. No rogue agents driving around unsupervised.

As organisations adopt M365 Copilot and Copilot Studio, agent sprawl becomes a real risk. Agent 365 governance means tracking all agents across Agent Builder and Copilot Studio, ensuring each has an owner, a purpose, and appropriate permissions. Copilot Studio agents promoted from Agent Builder inherit different governance — they move from M365 Copilot's simpler model to Power Platform's full governance framework.

✅ Admin Checklist

  • Maintain an agent registry (who owns each agent, its purpose, where it's deployed)
  • Set policies for agent creation — who can build agents and in which environments
  • Review agents promoted from Agent Builder to Copilot Studio — they may need new DLP compliance
  • Monitor agent credit consumption to prevent budget surprises
  • Establish a review process before agents are published to production
  • Archive or delete unused agents — they still consume environment resources
📄 Microsoft Learn docs →

📋 Compliance & Audit

💡 Compliance is like having a black box recorder in every car. If something goes wrong, you can replay exactly what happened, who was involved, and when. Audit logs, data residency, and sensitivity labels give you that accountability.

Copilot Studio supports enterprise compliance requirements including audit logging, geographic data residency, and integration with Microsoft Purview. Conversation transcripts can be retained, sensitivity labels respected, and all agent actions logged for regulatory review.

✅ Admin Checklist

  • Enable audit logging in the Power Platform admin centre
  • Verify data residency — Copilot Studio stores data in your tenant's geographic region
  • Configure Microsoft Purview sensitivity labels (Confidential/Highly Confidential files are not indexed)
  • Set data retention policies for conversation transcripts
  • Review Copilot Studio's compliance certifications (SOC 2, ISO 27001, GDPR, etc.)
  • For regulated industries: conduct a formal risk assessment before deploying customer-facing agents
  • Document your AI governance framework — include Copilot Studio agents in your responsible AI policy
📄 Microsoft Learn docs →

Design Patterns

Five proven architectures for Copilot Studio agents. Pick the one closest to your scenario and adapt.

Low complexity

Knowledge-grounded agent that answers questions from your documents. No topics needed.

When to use: When employees or customers need to ask questions about existing documentation — HR policies, product manuals, IT procedures.
Components needed: Knowledge sources (SharePoint/Web)Generative answersFallback topic

Architecture

┌─────────────┐     ┌──────────────────┐     ┌──────────────┐
│   User asks  │────▶│  Generative AI   │────▶│   Response   │
│   question   │     │  searches docs   │     │  with source │
└─────────────┘     └──────────────────┘     └──────────────┘
                           │
                    ┌──────┴──────┐
                    │  Knowledge  │
                    │  Sources    │
                    ├─────────────┤
                    │ SharePoint  │
                    │ Websites    │
                    │ Dataverse   │
                    └─────────────┘

How to Build It

  1. Create a new agent in Copilot Studio
  2. Add knowledge sources (SharePoint sites, uploaded files, or public URLs)
  3. Configure the system prompt with tone and scope instructions
  4. Customise the Fallback topic for unrecognised questions
  5. Test with real user questions before publishing
  6. Publish to Teams (internal) or web widget (external)

💡 Tips

  • Start with 1-2 knowledge sources and expand — don't dump everything in at once
  • Use the default GPT-4.1 mini model for FAQ bots — fast and cost-effective
  • Monitor analytics to find gaps — questions with no good answer reveal missing docs
Medium complexity

Agent that guides users through multi-step processes with Power Automate integration.

When to use: When conversations need to DO things — create tickets, send approvals, update records, trigger notifications.
Components needed: Topics with branchingPower Automate flowsAdaptive CardsVariablesDataverse

Architecture

┌──────────┐    ┌────────────┐    ┌───────────────┐    ┌──────────┐
│  User    │───▶│  Topic:    │───▶│ Power Automate│───▶│ Confirm  │
│  Request │    │  Collect   │    │ Flow: Create  │    │ to User  │
└──────────┘    │  Details   │    │ Ticket/Record │    └──────────┘
                └────────────┘    └───────────────┘
                     │                    │
                     ▼                    ▼
              ┌──────────────┐    ┌──────────────┐
              │ Adaptive Card│    │  Dataverse   │
              │ (input form) │    │  (store data)│
              └──────────────┘    └──────────────┘

How to Build It

  1. Map out the workflow on paper: what info is needed? What actions follow?
  2. Create a topic with trigger phrases for the workflow (e.g. 'submit expense')
  3. Add question nodes to collect required information from the user
  4. Use Adaptive Cards for structured data collection (forms, dropdowns)
  5. Create a Power Automate cloud flow for the backend action
  6. Connect the flow as a tool/action in the topic
  7. Add confirmation and error handling paths
  8. Store workflow state in Dataverse if multi-session

💡 Tips

  • Design the conversation flow before building it — whiteboard the happy path and error paths
  • Keep each topic focused on ONE workflow — don't combine ticket creation and approval in one topic
  • Use variables to carry data between conversation steps and Power Automate flows
Medium-High complexity

Agent that connects to external systems via connectors and custom APIs to provide live data.

When to use: When your agent needs real-time data from external systems — CRM records, inventory levels, pricing, or any REST API.
Components needed: Custom connectors or HTTP nodeAdaptive Cards (data display)TopicsGenerative orchestration

Architecture

┌──────────┐    ┌──────────────┐    ┌───────────────┐
│  User    │───▶│ Generative   │───▶│  Connector /  │
│  Query   │    │ Orchestrator │    │  HTTP Request  │
└──────────┘    └──────────────┘    └───────┬───────┘
                                            │
                    ┌───────────────────────┘
                    ▼
             ┌──────────────┐    ┌──────────────┐
             │ External API │    │ Parse & Show  │
             │ (CRM, ITSM,  │    │ Adaptive Card │
             │  Database)   │    │ to User       │
             └──────────────┘    └──────────────┘

How to Build It

  1. Identify the API your agent needs to call — get the endpoint URL, auth method, and schema
  2. Option A: Create a custom connector in Power Platform (reusable, manageable)
  3. Option B: Use the HTTP Request node for quick one-off API calls
  4. Define the tool/action with clear input parameters and output structure
  5. Use Adaptive Cards to display structured API responses (tables, lists, details)
  6. Handle API errors gracefully — show friendly messages, not raw error codes
  7. Test with real API calls including edge cases (no results, rate limits, timeouts)

💡 Tips

  • Custom connectors are better than raw HTTP for governance — admins can manage and DLP-control them
  • Cache frequently-requested data in Dataverse to reduce API calls and improve response time
  • Use generative orchestration with good tool descriptions — the AI will call the right API automatically
Medium complexity

One agent deployed across Teams (employees), website (customers), and mobile (WhatsApp) with channel-specific behaviour.

When to use: When different audiences need access to the same agent through different platforms — employees in Teams, customers on your website.
Components needed: Multi-channel publishingAuthentication per channelChannel-specific topicsWeb widget customisation

Architecture

                    ┌──────────────────┐
                    │   Copilot Studio │
                    │   (ONE agent)    │
                    └────────┬─────────┘
                             │
            ┌────────────────┼────────────────┐
            ▼                ▼                ▼
     ┌────────────┐   ┌────────────┐   ┌────────────┐
     │   Teams    │   │  Website   │   │  WhatsApp  │
     │  (SSO)     │   │  (widget)  │   │  (Meta)    │
     │ Employees  │   │  Customers │   │  Mobile    │
     └────────────┘   └────────────┘   └────────────┘

How to Build It

  1. Build and test your agent in Copilot Studio first — get the core working
  2. Configure authentication for each channel: SSO for Teams, manual auth or none for web
  3. Publish to Teams via the org app catalogue (may need admin approval)
  4. Generate the web widget embed code and add to your website
  5. For WhatsApp: register a WhatsApp Business account and connect via the channel settings
  6. Use channel-specific topic conditions if behaviour should differ per channel
  7. Test in each channel — Adaptive Card rendering varies across platforms

💡 Tips

  • Start with Teams (easiest), then add web, then mobile channels
  • External channels can drive unexpected credit consumption — set up monitoring
  • Web channel security prevents your widget from being embedded on unauthorised sites
Low-Medium complexity

Migrating an agent from Agent Builder to Copilot Studio for advanced capabilities.

When to use: When your Agent Builder agent has outgrown its limits — you need topics, connectors, workflows, multi-channel, or governance.
Components needed: Agent Builder source agentCopilot Studio destinationReconfiguration of advanced features

Architecture

┌─────────────────┐         ┌─────────────────────┐
│  Agent Builder   │  Copy   │   Copilot Studio    │
│  ┌────────────┐ │ ──────▶ │  ┌────────────────┐ │
│  │Instructions│ │  ✅     │  │ Instructions   │ │
│  │Knowledge   │ │  ✅     │  │ Knowledge      │ │
│  │Basic config│ │  ✅     │  │ + Topics       │ │
│  └────────────┘ │         │  │ + Connectors   │ │
│                 │         │  │ + Channels     │ │
│  Limits:        │         │  │ + Power Auto   │ │
│  • 8K chars     │         │  │ + Governance   │ │
│  • No topics    │         │  └────────────────┘ │
│  • M365 only    │         │                     │
└─────────────────┘         └─────────────────────┘

How to Build It

  1. In Agent Builder, open your agent and select 'Edit in Copilot Studio'
  2. This copies: instructions, knowledge sources, basic configuration
  3. Review and update the instruction text (Studio allows 16K chars vs Builder's 8K)
  4. Add topics for any multi-step conversations you couldn't build before
  5. Add connectors and Power Automate flows for external system integration
  6. Configure additional channels beyond Teams
  7. Set up environment governance (DLP, RBAC) for the new agent
  8. Test thoroughly — behaviour may differ slightly with the richer orchestration engine
  9. Note: This is a one-way copy. The Agent Builder version remains but changes won't sync back.

✅ What Carries Over

  • Instructions (system prompt)
  • Knowledge sources (SharePoint, files, URLs)
  • Basic agent configuration and name

🔨 What Needs Rebuilding

  • No topics exist — you build them fresh in Studio
  • No connectors — add them in the Studio environment
  • No Power Automate flows — create and connect them
  • Authentication may need reconfiguration for new channels
  • DLP policies in the target environment may restrict some features

💡 Tips

  • Don't migrate just because you can — only promote when you genuinely need Studio's features
  • The Agent Builder version continues to work independently after the copy
  • Test the Studio version with real users before retiring the Agent Builder version

🚫 Anti-Patterns — What NOT to Do

Learn from common mistakes. These patterns cause projects to fail.

🏗️ The Monolith Agent

One giant agent that does everything — HR, IT, Finance, Sales — with 50+ topics and every connector enabled.

Why it fails: Impossible to maintain, test, or govern. DLP becomes a nightmare because you can't restrict connectors per department. Performance degrades with too many topics competing for intent matching.
✅ Instead: Build specialised agents per department. Use multi-agent orchestration to connect them.
🚫 Ship It Without Governance

Building and publishing agents directly in the default environment with no DLP, no RBAC, no approval process.

Why it fails: Shadow AI. No audit trail. Agents accessing sensitive data without controls. One DLP change breaks everything. Compliance nightmare.
✅ Instead: Set up Dev/Test/Prod environments BEFORE anyone builds. Apply DLP policies. Require admin approval for publishing.
🎰 Skip Testing — It Worked in Preview

Testing only in the built-in test chat, then publishing straight to production.

Why it fails: Test chat doesn't replicate real-world conditions — different channels render Adaptive Cards differently, auth flows differ, concurrent users expose race conditions.
✅ Instead: Test in every target channel. Use Agent Evaluations for automated testing. Deploy to Test environment first. Get real users to UAT.
💸 Ignore Credit Consumption

Building a complex agent with GenAI, tools, autonomous triggers, and external channels — without estimating credit usage.

Why it fails: A single complex interaction can burn 20+ credits. External-facing agents can spike unpredictably. You'll blow through 25K credits in a week.
✅ Instead: Use the Credit Estimator (Tab 6). Set up consumption monitoring. Add pay-as-you-go as a safety net. Start with simple patterns and add complexity.
🔨 Studio for Simple Q&A

Using Copilot Studio with topics, connectors, and custom flows for a simple HR FAQ that just answers questions from SharePoint documents.

Why it fails: Unnecessary complexity. Costs credits when Agent Builder is free with M365 Copilot. Overkill maintenance burden.
✅ Instead: Start with Agent Builder. It handles document Q&A perfectly. Only promote to Studio when you genuinely need topics, connectors, or multi-channel.

Agent Planner

Answer 5 questions to generate a build plan for your Copilot Studio agent.

1 What's the agent's purpose?
2 Who will use this agent?
3 Where should it be available?
4 What systems does it need to connect to?
5 How complex is the conversation?

What's the agent's purpose?

Who will use this agent?

Where should it be available?

What systems does it need to connect to?

How complex is the conversation?

💰 Credit Estimator

Estimate your monthly Copilot Studio credit consumption.

100
60%
20%
0
0 credits/month

Frequently Asked Questions

What's the difference between Agent Builder and Copilot Studio?

Agent Builder is the no-code, instruction-based agent creator built into M365 Copilot — great for simple Q&A agents grounded in your M365 data. Copilot Studio is the full low-code platform with conversation topics, 1,000+ connectors, Power Automate integration, multi-channel deployment, and enterprise governance. Think of Agent Builder as a microwave (quick and easy) and Copilot Studio as a full kitchen (more capability, more control). You can promote agents from Builder to Studio with one click when they outgrow Builder’s limits.

Do I need a separate licence for Copilot Studio?

Yes. Copilot Studio requires its own tenant licence — a prepaid pack of 25,000 Copilot Credits per month, or pay-as-you-go at $0.01 per credit via Azure. Agent Builder, by contrast, is included with your M365 Copilot licence at no extra cost. If you only have M365 Copilot users, they get limited Copilot Studio access for agents shared with them, but building advanced agents requires the Studio licence.

What are Copilot Credits and how fast do they run out?

Copilot Credits replaced ‘messages’ in September 2025. A regular topic-based response costs 1 credit. A generative AI answer costs 2 credits. Complex interactions with tools, flows, and autonomous triggers can use 20+ credits per turn. With 25,000 credits per month, a busy FAQ bot handling 200 queries/day uses about 12,000 credits/month (at 2 credits each). Monitor usage carefully — add pay-as-you-go as a safety net.

Can I deploy a Copilot Studio agent to my website for external customers?

Yes. Copilot Studio supports multi-channel deployment including web widgets, WhatsApp, Facebook, and custom channels. For external customers, you configure unauthenticated access or manual authentication (rather than SSO). Enable web channel security to prevent your widget being embedded on unauthorised sites. External-facing agents can drive unpredictable credit consumption, so set up monitoring and alerts.

How do DLP policies affect my agents?

Data Loss Prevention policies control which connectors and channels your agents can use. Since early 2025, DLP enforcement is mandatory for all tenants — no exemptions. Admins configure DLP in the Power Platform admin centre, classifying connectors into Business, Non-business, or Blocked groups. If a connector your agent needs is blocked by DLP, you’ll see a detailed error message naming the policy. Work with your admin to adjust the policy or choose an alternative connector.

Should I use Copilot Studio or Azure AI Foundry?

Use Copilot Studio when: you’re an IT admin or citizen developer, you want low-code/no-code, you need Power Platform integration, and your agents serve internal employees or standard customer support. Use Azure AI Foundry when: you have developers writing Python/C#, you need custom model training or fine-tuning, you require sovereign/air-gapped deployment, or you’re building a customer-facing AI product at massive scale. Most organisations start with Copilot Studio and only go to Foundry when they hit its limits.

Can I connect Copilot Studio agents to Azure AI Foundry?

Yes. Copilot Studio supports multi-agent orchestration where a Studio agent can call an AI Foundry agent as a connected agent. This hybrid pattern lets you keep the low-code authoring experience of Copilot Studio while leveraging AI Foundry’s custom models for specific tasks. It’s the best of both worlds for organisations that need some custom AI without going all-in on pro-code.

What governance controls should I set up before letting people build agents?

Start with the basics: (1) Create separate Dev/Test/Production environments, (2) Apply DLP policies that block unnecessary connectors and channels, (3) Control who can create agents using Entra ID security groups, (4) Require admin approval before agents are published to the org app catalogue, (5) Enable audit logging, (6) Set up credit consumption monitoring. Use the Power Platform Centre of Excellence Starter Kit for cross-tenant visibility into all agents.

📝 What's New in Copilot Studio

April 2026
  • Computer Use now GA — agents can control desktop apps via Claude Sonnet 4.5
  • A2A (Agent-to-Agent) protocol support — connect to agents on any platform
  • Grok 3 model added to model catalogue
  • MCP server connections now GA — extend agents with external tool servers
  • Generative builder improvements — describe what you want, AI builds the topic
  • SharePoint channel now available for embedding agents on SP pages
March 2026
  • Claude Opus 4 added as Deep reasoning model option
  • Agent evaluations GA — automated testing before deployment
  • Enhanced analytics with conversation transcripts and topic performance
  • Environment routing improvements — new makers auto-directed to designated environments
February 2026
  • GPT-5 added to General model tier
  • Multi-agent orchestration improvements — better child agent handoff
  • DLP endpoint filtering for HTTP connector — allow specific URLs instead of blocking all
  • Improved Adaptive Card designer in the authoring canvas
January 2026
  • Auto model selection (per-turn adaptive routing) now GA
  • Autonomous agent triggers expanded — Dataverse, HTTP, scheduled
  • Power Automate integration improvements — async flow patterns supported
  • Content moderation per-prompt configuration
Last updated: April 2026 · Sources: Microsoft Learn, Power Platform admin centre

🤖 Your Copilot Decision Toolkit

🎯Readiness

Are you ready?

📊Features

What can it do?

💰ROI

Is it worth it?

📜Licensing

What licence?

🛒Licence Picker

Cheapest combo

🎬Demo Scripts

Show the value

⚙️Studio

Build agents

💬 Got feedback? Share it here →

💬