TL;DR: Copilot Studio now supports two powerful new workflow nodes — Extract Node pulls structured data from documents like invoices and contracts, while Copilot Node lets your agent delegate work to Microsoft 365 Copilot directly within a flow. Together they enable truly autonomous agents that read, reason, and act.
A few months ago, building an autonomous agent meant stitching together multiple Power Automate flows, custom connectors, and AI Builder models — a task that required a developer familiar with all three tools.
That complexity has come down significantly.
Copilot Studio’s latest wave of updates introduces two workflow nodes that shift agent-building from a multi-tool exercise into something a skilled business analyst or citizen developer can accomplish in a single designer session.
If you are a Power Platform maker, a Copilot Studio developer, or a consultant helping clients automate document-heavy business processes, here is what the Extract Node and Copilot Node do — and where they fit in a real autonomous agent workflow.
Official Website: Microsoft Copilot Studio Blog
What is generative orchestration in Copilot Studio?
What makes an agent “autonomous”?
Traditional chatbots respond to questions. Autonomous agents complete objectives.
The difference is initiative and persistence. An autonomous agent with generative orchestration can:
- Receive a goal (e.g., “Process this invoice and notify the finance team if it exceeds the approved vendor budget”)
- Plan the steps needed to complete it
- Invoke tools, nodes, and other agents in sequence
- Handle variations and edge cases using AI reasoning rather than rigid branching rules
Copilot Studio’s generative orchestration engine powers this. The Extract and Copilot nodes are the building blocks that give agents the tools to act on real-world business data.
1. The Extract Node: Reading Documents Automatically
What does the Extract Node do?
The Extract Node transforms unstructured document content into structured data that your workflow can use.
In plain terms: instead of your agent receiving a PDF invoice and not knowing what to do with it, the Extract Node reads the document, pulls out the fields you care about (vendor name, invoice total, line items, due date), and makes those values available as dynamic content for the next steps in your flow.
What document types does the Extract Node support?
- PDF files
- Microsoft Word documents (.docx)
- Microsoft Excel spreadsheets (.xlsx)
- Microsoft PowerPoint presentations (.pptx)
This covers most common business document formats — invoices, contracts, financial statements, approval forms, and reports.
How do you set up the Extract Node?
- Open Copilot Studio and navigate to your agent’s workflow.
- Add a new node → select Extract Node from the node type menu.
- Upload a sample document that represents the type of document your agent will process (e.g., a sample invoice from a vendor).
- Define the extraction schema — Copilot Studio analyses the sample and suggests field mappings automatically. You confirm, rename, or add fields to the schema.
- Test inline — The node provides a testing panel where you can drop in another document to verify the extraction accuracy before saving.
- Connect the output — Extracted field values become dynamic content tokens you can reference in subsequent nodes (e.g.,
Invoice.TotalAmount,Invoice.VendorName).
Real example
A vendor sends an invoice PDF by email. An agent receives the email attachment, passes it to the Extract Node, and extracts:
VendorName: Contoso LtdInvoiceNumber: INV-2026-08812TotalAmount: £14,750DueDate: 2026-10-31
The next step checks TotalAmount against an approved vendor budget in a SharePoint list. If it exceeds the threshold, the agent routes it to a finance manager for approval. If it is within budget, it posts a summary to the procurement Teams channel and marks the invoice as received.
All without a human manually opening the PDF.
2. The Copilot Node: Delegating Work to Microsoft 365 Copilot
What does the Copilot Node do?
The Copilot Node lets your Copilot Studio workflow hand off a task to Microsoft 365 Copilot — and get a result back to use in subsequent steps.
This is different from simply calling a prompt. The Copilot Node can invoke Copilot Chat for quick analysis or Cowork for longer, multi-step delegated tasks. It grounds the request in the connected user’s Microsoft 365 data — their emails, files, calendar, and Teams chats.
What are the two modes of the Copilot Node?
Chat mode: Use this for focused, single-round interactions with Copilot. Good for:
- Generating a short summary of extracted data
- Drafting a notification message with the extracted values filled in
- Calling a specific Microsoft-built agent (Researcher, Analyst) for additional context
- Calling a custom agent built in Agent Builder
Cowork mode: Use this for delegating complex work that spans multiple apps and data sources. Good for:
- Asking Copilot to draft an approval email incorporating the vendor’s invoice details and the budget comparison result
- Requesting a calendar summary ahead of a follow-up meeting with the vendor
- Instructing Copilot to resume an existing Cowork task from an earlier stage
Both modes accept dynamic content tokens from earlier workflow steps, so the Copilot prompt is always personalised to the actual data in the current run.
Real example (continuing the invoice scenario)
After the Extract Node pulls the invoice details and a budget check confirms the amount is within limits, the Copilot Node runs a Chat task:
Instruction: “Draft a confirmation email to [Invoice.VendorName] confirming receipt of invoice [Invoice.InvoiceNumber] for [Invoice.TotalAmount], due [Invoice.DueDate]. Keep it professional and under 100 words.”
Copilot generates the email draft and returns it as the Copilot Node’s output. The next step sends that draft via Outlook — processed, contextualised, and sent — with no human typing a single word.
What else is in this Copilot Studio wave?
Content Safety Tests (Now GA)
Microsoft moved configurable content safety tests to General Availability in September 2026. Every agent response can now be automatically evaluated against Microsoft’s responsible AI content policies before it is delivered to users.
Administrators can configure severity thresholds per agent, balancing safety strictness against operational needs.
Model Selection via GitHub Copilot Harness
Copilot Studio agents can now access experimental and frontier AI models through a model dropdown in the agent designer, powered by the GitHub Copilot harness. This lets developers evaluate how different models perform on their specific agent tasks before committing to a production deployment.
Note: agents using frontier models through the GitHub Copilot harness are billed using Copilot Credits (usage-based), not the flat seat license.
Building Autonomous Agents — Summary
| Node / Feature | What it does | Best used for |
|---|---|---|
| Extract Node | Pulls structured field data from PDFs, Word, Excel, PowerPoint | Invoice processing, contract extraction, form data ingestion |
| Copilot Node (Chat) | Sends a grounded prompt to Copilot Chat and returns the result | Summaries, draft generation, calling specific agents |
| Copilot Node (Cowork) | Delegates a complex multi-step task to Copilot and retrieves the outcome | Email drafting, cross-app coordination, resumed Cowork tasks |
| Content Safety Tests | Evaluates agent responses for harmful content before delivery (GA) | Enterprise agent governance and responsible AI compliance |
| Model Selector | Choose AI models via GitHub Copilot harness in the agent designer | Evaluating experimental models before production rollout |
What should developers and makers do next?
Three practical starting points:
- Build the invoice extraction scenario — Create a new agent in Copilot Studio, add an Extract Node with a sample invoice, define a schema for five key fields, and test the extraction. This is a 30-minute exercise that demonstrates the full capability.
- Combine Extract and Copilot nodes in one flow — Extend the extraction scenario by adding a Copilot Node in Chat mode to draft a summary notification using the extracted values. This end-to-end pattern applies to dozens of document-heavy business processes.
- Assign Copilot Credits before enabling frontier model agents — If your agents use the GitHub Copilot harness and frontier models, ensure spending policies are configured in the Microsoft 365 Admin Center before deploying to production users.
These nodes represent a meaningful maturation in Copilot Studio’s agent capabilities. Start with a contained use case, validate the extraction accuracy, and expand from there.
FAQs: Copilot Studio Autonomous Agents
Does the Extract Node require AI Builder credits?
No. The Extract Node in Copilot Studio workflows is powered by the agent’s own AI layer, not AI Builder. You do not need separate AI Builder credits to use the Extract Node within Copilot Studio.
Can the Copilot Node access files from SharePoint and OneDrive?
Yes, when running in Cowork mode, the Copilot Node is grounded in the connected user’s Microsoft 365 data — including files in SharePoint and OneDrive, emails, calendar, and Teams chats — provided the user has consented to that grounding.
What happens if the Extract Node cannot parse a field from an unusual document?
The Extract Node returns a null value for fields it cannot confidently extract. Your workflow should include conditional branching to handle null extractions — for example, routing the document to a human reviewer if a critical field like TotalAmount is missing.
Can the Copilot Node call a custom agent I built in Agent Builder?
Yes. In Chat mode, the Copilot Node can invoke specific Microsoft-built agents (Researcher, Analyst) or custom agents you have created in Agent Builder. The custom agent must be published and accessible to the connected user.
Are agents using the Copilot Node billed under my seat license or Copilot Credits?
Standard Copilot Chat calls within the Copilot Node are covered by the seat license. Frontier model sessions and Cowork tasks draw from Copilot Credits (usage-based billing). Review your spending policies before deploying agents that use Cowork mode at scale.
Is generative orchestration available for all Copilot Studio plans?
Generative orchestration requires a Microsoft 365 Copilot license or the Copilot Studio standalone license with the generative AI add-on. Check Microsoft’s current licensing page for your specific plan’s eligibility.
