Key takeaways
- Copilot agents are focused assistants inside Microsoft 365 with their own instructions, knowledge and actions.
- They do not retrain themselves from conversations; they improve when owners update instructions and knowledge.
- Agents respect each user's existing permissions, and run in Microsoft's cloud within your organisation's tenant.
- Usage may consume Copilot Studio credits depending on licences and what the agent does.
- SharePoint agents answer questions grounded in a specific site, library or set of files.
Copilot agents are focused AI assistants inside Microsoft 365 Copilot, each with its own instructions, knowledge sources and actions for a specific job, such as answering HR policy questions, helping new hires, or handling IT requests. Microsoft provides some built-in agents, and organisations build their own with Agent Builder or Copilot Studio. This page explains what agents are used for, when to use one, whether they learn or remember, how credits apply, privacy and where they run, how good they are in practice, and how agents work in SharePoint.
What are Copilot agents?
Copilot agents are specialised assistants built on Microsoft 365 Copilot. Each one combines instructions that define its role, knowledge it can search, and optionally actions it can take in other systems, so it handles one kind of task better than general Copilot.
There are a few kinds. Retrieval agents answer from specific knowledge, such as a policy site. Task agents take actions, such as creating a ticket or updating a record. Autonomous agents run on triggers or schedules without a user starting a chat. Microsoft also ships prebuilt agents, such as research and analysis agents, and agents in Teams and SharePoint. You find agents in the Copilot app or Teams, and you can mention them in a chat to call them in.
What are Copilot agents used for?
Organisations use Copilot agents to answer recurring questions, guide people through processes and automate routine tasks: HR and policy questions, employee onboarding, IT helpdesk, sales preparation, project status and customer support.
Examples that work well:
- An onboarding agent that answers new joiners' questions from the handbook and links the right forms.
- An IT agent that resolves common issues and opens a ticket when it cannot.
- A sales agent that summarises an account from CRM data before a call.
- A project agent that reports status from a SharePoint site and task lists.
- A college department agent that answers questions about syllabus, deadlines and lab rules.
The common pattern is a clear audience, trusted sources and a defined hand-off to a person.
When should I use a Copilot agent?
Use an agent when the same kind of question or task keeps coming up, the answers live in known sources, and a focused assistant would be more accurate than general Copilot. Use general Copilot for one-off, varied requests.
Signals an agent is worth building: a team answers the same twenty questions every week; a process has steps people often get wrong; information is scattered across sites and documents; or a task needs data from another system. Signals it is not: the questions are rare or completely different each time, the source documents are outdated or contradictory, or nobody will own and maintain the agent. Clean, current knowledge matters more than clever instructions.
Do Copilot agents learn?
Not in the sense of retraining themselves. Agents do not update the underlying model from your conversations. They improve when their owners update instructions, add or fix knowledge sources, and refine actions based on what users ask.
That is by design: a model that changed itself from every conversation would be unpredictable and could absorb wrong or sensitive information. What looks like learning comes from fresh knowledge: when a policy document is updated in SharePoint, the agent's next answer reflects it. Owners should review conversation analytics regularly, note questions the agent answered poorly, and fix the source content or instructions. That loop is the real way agents get better over time.
Do Copilot agents have memory?
Within a conversation, yes: an agent remembers what has been said in that chat. Across conversations, memory is limited and depends on the agent and Microsoft's features; agents generally do not carry details from one user's chats into another's.
Microsoft 365 Copilot has been adding personalisation and memory features for individual users, controlled by user and admin settings. Agents built in Copilot Studio can store and use variables during a conversation and can save information to systems such as Dataverse or a SharePoint list through actions, which acts as explicit, controlled memory. If an agent must remember something, such as a user's preferences or a case history, store it deliberately in a system of record rather than relying on chat history.
Do Copilot agents use credits?
It depends on the agent and licensing. Agents built in Copilot Studio consume Copilot Credits for their interactions and actions, especially when used by people without a Microsoft 365 Copilot licence, on external channels or when running autonomously.
Also asked as: when to use copilot agents
Users with a Microsoft 365 Copilot licence can use many agents within Microsoft 365 as part of that licence, while organisations without it pay for agent usage through Copilot Studio capacity or pay-as-you-go billing. Actions that call generative AI, search knowledge or run autonomously typically consume more. Microsoft updates these rules, so admins should check the current Copilot Studio licensing guide and monitor consumption in the Power Platform admin centre.
Are Copilot agents private?
Agents follow Microsoft 365's security model: they only access content that the user, or the configured identity for autonomous agents, is allowed to see, and data stays within your organisation's Microsoft 365 environment under enterprise data protection.
Privacy risks come from configuration, not the platform. An agent with an uploaded file shares that file's content with everyone who can use the agent. An agent using a service account can expose data the end user should not see. SharePoint sites with overly broad permissions make agents surface documents widely. Before sharing an agent, check its knowledge and actions against its audience, and ask your admin about policies for agent creation and sharing.
Where do Copilot agents run?
Copilot agents run in Microsoft's cloud, within your organisation's Microsoft 365 tenant and Power Platform environments. You use them through Microsoft 365 Copilot, Teams, SharePoint, websites or other channels where they are published.
Copilot Studio agents live in a Power Platform environment, which sets the region where their data is stored. Agent Builder agents are stored with Microsoft 365. The language models run in Microsoft's data centres under the same commitments as Microsoft 365 Copilot. Nothing runs on your laptop, except the app or browser you use to chat. For data residency questions, admins can check environment regions and Microsoft's data boundary commitments for their geography.
Are Copilot agents any good?
They are good when the scope is narrow, the knowledge is clean and current, and someone owns them. They disappoint when built as general "ask me anything" bots over messy, outdated document collections.
What separates useful agents from poor ones:
- A specific audience and purpose written into the instructions.
- A small set of authoritative sources, with duplicates and old versions removed.
- Clear behaviour when the answer is not found: say so and hand over.
- Tested with real questions before launch.
- Reviewed regularly using analytics and feedback.
The same lesson shaped RAG.NextUpgrad: retrieval quality and a refusal path when evidence is weak matter more than the model. An honest "I do not know" beats a confident wrong answer.
How do Copilot agents work in SharePoint?
SharePoint agents answer questions grounded in the content of a specific site, document library, folder or set of files. Every site can offer a ready-made agent, and site members can create custom agents scoped to chosen content.
To create one, select the files or library, choose to create an agent, give it a name, instructions and starter prompts, and save it; it becomes an item in the library that you can share and use in Teams chats. The agent searches only its scoped content, within each user's permissions, and cites the documents it used. This works well for project sites, policy libraries and course material. Keep the scoped content tidy: archive old versions and use clear titles so the agent finds the right document.
Sources
GenAI engineer and AI product builder with 2+ years shipping production-grade AI systems: RAG pipelines, fine-tuned LLMs, hybrid retrieval and multi-modal apps across vision, speech and OCR, architected end to end from ingestion to deployment. Leads engineering for SCULT INDIA's 14-member team, founded the 500+ member TechVerse Enclave community and has mentored 200+ students. Three hackathon first prizes: Changethon 2025 (IIT Roorkee), Product Genesis at Vividhotsava 2025 (CSJMU) and BYTEBATTLE (MeetKats).
Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges. For campus sessions the inviting institution covers the visit end to end: travel (a First AC train ticket for journeys up to five hours, a flight for anything longer), accommodation, meals, and pickup and drop; the honorarium and workshop fees are negotiable on WhatsApp or on the call, and requests are welcome from institutions that can cover the visit this way. Email pranjulrathour41@gmail.com.
