Copilot Studio: Lite vs Full

Since late 2025, Microsoft offers two tiers. Understanding which to use saves licensing headaches.

How to Access Copilot Studio

Build Your First Agent

Create an agent in minutes using natural language — no code required.

Creating the Agent

The Instructions (System Prompt)

The Instructions field is essentially the agent's system prompt — it defines personality, behaviour, and constraints. This can be up to 8,000 characters. Good instructions include:

• Role — "You are a helpful HR assistant for Sustainsys Consulting"
• Tone — "Be professional but friendly, use UK English"
• Boundaries — "Only answer questions about HR policies. For other topics, redirect to the appropriate team"
• Format — "Keep answers concise. Use bullet points for lists. Always cite the source document"

Test As You Build

Copilot Studio has a live test panel on the right side. After every change, click "Start new test session" and try asking questions. This is the fastest feedback loop — change instructions, test, refine, repeat.

Connect Knowledge Sources

Knowledge sources are what make your agent smart about your specific business data. Without them, the agent only has generic AI knowledge.

Available Knowledge Types

Adding Knowledge

How It Works Under the Hood

When you add knowledge, Copilot Studio:

• Chunks uploaded files into smaller pieces for faster processing
• Creates vector embeddings for semantic search (stored in Dataverse)
• When a user asks a question, it finds the most relevant chunks that match
• Feeds those chunks to the LLM as context for generating the response
• Returns the answer with citations back to the source

This is essentially a managed RAG pipeline — the same architecture you've been building with ChromaDB and LlamaIndex, but handled by Microsoft's infrastructure.

Topics & Conversation Flow

Topics are the building blocks of your agent's behaviour — think of them as "intents" with built-in conversation logic.

What Are Topics?

A topic is a discrete conversation path your agent can follow. It has:

• Trigger phrases — what the user says to activate it (e.g., "book a meeting room", "reset my password")
• Conversation nodes — the steps: ask questions, show messages, call actions, branch based on conditions
• Variables — store user input and pass data between nodes

Two Orchestration Modes

Creating a Topic

Actions, Tools & Integrations

Actions turn your agent from a Q&A bot into something that can actually do things — send emails, create tickets, update records.

What Can Agents Do?

Adding an Action

Publishing Your Agent

Once your agent is ready, you can deploy it to multiple channels:

• Microsoft Teams — appears as a chat bot in Teams
• Microsoft 365 Copilot — accessible via @mention in Copilot
• SharePoint — embedded on intranet pages
• Websites — embed via iframe or Web Chat widget (Full tier)
• WhatsApp — connect via Twilio or direct (Full tier)
• Demo website — quick shareable link for testing

Understanding Prompts in Microsoft's Ecosystem

Microsoft has two related but distinct "prompt" concepts. Let's untangle them.

Prompt Gallery (M365 Copilot)

The Prompt Gallery lives inside Microsoft 365 Copilot (the chat experience). It's a catalogue of ready-made prompts that help users get more out of Copilot. Think of it as a recipe book for Copilot users.

• Microsoft-authored prompts — pre-built for Word, Excel, PowerPoint, Outlook, Teams
• User-created prompts — you can save your own effective prompts
• Shared prompts — share with your team or organisation via collections
• Stored securely in Microsoft 365 Substrate data store

Access it at m365.cloud.microsoft/copilot-prompts or within M365 Copilot.

Prompt Builder (Copilot Studio)

The Prompt Builder in Copilot Studio is more powerful — it lets you create reusable, modular AI prompts that guide agents to perform specific tasks. These are essentially prompt templates with dynamic inputs.

• Create prompts for: summarising, classifying, extracting, translating, sentiment analysis
• Use input variables for dynamic context at runtime
• Add knowledge data to ground responses
• Use Power Fx logic inside prompt inputs for calculations and formatting
• Share prompts with other makers and reuse across agents and flows

Build a Prompt Library

A structured collection of reusable prompts that your organisation can share, standardise, and continuously improve.

Organising Your Library

A good prompt library is organised by function, not by tool. Group prompts into categories your teams actually use:

Creating a Prompt in Copilot Studio

You are a document summariser for Sustainsys Consulting.

Given the following document content:
[DocumentContent]

Provide a summary that includes:
- Key findings (3-5 bullet points)
- Risks identified
- Recommended next steps

Keep the summary under 300 words. Use UK English.
Format as markdown with clear headings.

Sharing via M365 Prompt Gallery

For end-user prompts (not agent-builder prompts), use the M365 Prompt Gallery:

• Create prompts in Microsoft 365 Copilot and save them
• Click "Share" to share with specific teams or security groups
• Shared prompts appear in colleagues' Prompt Gallery under "Shared with me"
• Admins can track usage via Copilot analytics

Reusable Knowledge Packs

Combine curated knowledge sources + tested prompts + agent instructions into reusable packages that can be deployed across multiple agents.

What is a Knowledge Pack?

A knowledge pack is a reusable bundle that packages together everything an agent needs to handle a specific domain:

Building a Knowledge Pack

Here's a practical pattern for creating a reusable knowledge pack using the tools available today:

1. Curate Your Documents

Gather the authoritative documents for your domain into a dedicated SharePoint folder or document library. Keep it focused — an HR knowledge pack should only have HR documents, not the entire company wiki.

2. Create a Glossary

Build a glossary of domain-specific terms. In Copilot Studio, this can be a CSV file in Dataverse with columns for term, definition, and context. This helps the agent correctly interpret industry jargon — especially important in fields like financial services and healthcare.

Term,Definition,Context
BAU,Business As Usual,Operations and project management
SLA,Service Level Agreement,Client contracts and service delivery
RAG,Red Amber Green status,Project reporting (not Retrieval Augmented Generation)
MoSCoW,Must Should Could Won't,Requirements prioritisation method

3. Write Domain-Specific Instructions

Create a set of agent instructions tailored to this domain. These can be saved and reused across multiple agents:

# HR Policy Assistant Instructions

You are an HR policy assistant for Sustainsys Consulting Ltd.

## Behaviour Rules
- Always cite the specific policy document and section
- If a policy has changed recently, flag that it may have updated
- For sensitive topics (disciplinary, redundancy), suggest speaking
  to HR directly rather than relying solely on policy documents
- Use UK employment law terminology

## Response Format
- Lead with a direct answer
- Follow with the relevant policy details
- End with "Source: [document name], Section [X]"

## Glossary Reference
- Refer to {Global.Glossary} for term definitions
- When users use acronyms, expand them on first use

4. Bundle Prompt Templates

Create prompts specific to this domain and save them in the Prompt Library:

• "Summarise this HR policy for a new starter" (with variable for policy document)
• "Compare two policy versions and highlight what changed"
• "Draft a response to this employee query about [topic]"

5. Package as a Copilot Studio Solution

In Power Platform, you can export your agent (with all its knowledge, topics, and prompts) as a solution. This solution can be imported into other environments — making it a true reusable knowledge pack.

Multi-Agent Orchestration & MCP

For complex business processes, one agent isn't enough. Copilot Studio supports routing tasks to specialised agents.

Multi-Agent Orchestration

With multi-agent orchestration, you can build a system where a primary agent routes tasks to specialised agents based on the user's intent:

Each specialist agent has its own knowledge pack, instructions, and tools. The orchestrator decides which specialist to invoke based on the user's query.

MCP in Copilot Studio

Copilot Studio now supports Model Context Protocol (MCP) — the same standard you've been working with in your LangGraph projects. This means:

• You can connect MCP servers as tools for your agents
• Agents can discover and use MCP-exposed capabilities at runtime
• Your custom MCP servers (like the ones you built for Gmail/weather) could potentially be connected to Copilot Studio agents
• This bridges the gap between your Python/LangGraph agent work and Microsoft's low-code platform

What's Next?

You now understand the full landscape of Microsoft's agent platform and prompt ecosystem.

Hands-On Next Steps

Consulting Opportunities

These skills map directly to consulting engagements you could offer through Sustainsys:

• Agent Assessment — audit a client's M365 environment and identify high-value agent use cases
• Knowledge Pack Development — build domain-specific knowledge packs for financial services, healthcare, biosciences
• Prompt Library Strategy — design and deploy organisation-wide prompt libraries with governance
• Custom Agent Development — build sophisticated agents with Power Automate integration
• Bridge Building — connect custom Python/LangGraph solutions to Copilot Studio via MCP

Key Resources

• Copilot Studio documentation
• Agent adoption resources
• Copilot Studio Kit — testing, governance, analytics tools
• Prompts overview
• Skills for Copilot Studio (community)