Power Automate 2025 Wave 2: Every New Feature You Need to Know

Wave 2 at a Glance

Microsoft's Power Automate 2025 Wave 2 release plan covers features shipping between October 2025 and March 2026. The wave is the largest AI-focused update Power Automate has received — nearly every product area gained a Copilot-assisted capability, and several long-standing pain points in desktop automation and document processing were addressed with substantial engineering investment.

Key milestones by general availability (GA) date:

• October 2025: Copilot-powered flow building in cloud flows, Desktop Flow Recorder v2 GA, AI Builder prebuilt document models refresh.
• November 2025: New SharePoint and Teams connector actions GA, Process Mining bottleneck detection GA.
• December 2025: SAP connector updates, Process Advisor enhanced variance analysis.
• January – March 2026: AI agent flows preview, natural language approvals, expanded Copilot in desktop flows.

Copilot-Powered Flow Building

The headline feature of Wave 2 is the ability to describe a flow in plain English and have Copilot generate the full flow definition — trigger, actions, conditions, and expressions — ready to review and save. This is a step beyond the Wave 1 "Describe it to design it" capability, which only generated a starting trigger and the first action.

Wave 2 Copilot understands multi-step flows with branching logic. A prompt like "When a new item is added to a SharePoint list, check if the Status column is Approved, then send an email to the item creator and post a Teams message to the Approvals channel" now generates the complete flow, including the triggerOutputs() expressions for the SharePoint trigger and the correctly scoped Teams connector action.

What Copilot Generates Well

• Standard connector combinations (SharePoint, Teams, Outlook, Dataverse).
• Approval flows with Respond to an approval action wired correctly.
• Scheduled flows with recurrence triggers and condition branches.
• Simple HTTP request flows calling external APIs with static headers.

Where You Still Need to Intervene

Copilot struggles with flows requiring complex dynamic expressions, loops over arrays, or pagination patterns for large SharePoint list queries. Treat Copilot output as a 70% starting point — always review expressions and test with real data before deploying to production.

// Expression Copilot generates for "item creator email" from SharePoint trigger
triggerOutputs()?['body/Author/Email']

// Copilot-generated condition to check Status column value
@equals(triggerOutputs()?['body/Status/Value'], 'Approved')

// Dynamic Teams channel message body Copilot inserts
concat(
  'New approved item: ',
  triggerOutputs()?['body/Title'],
  ' — submitted by ',
  triggerOutputs()?['body/Author/DisplayName']
)

AI Builder Document Processing Improvements

AI Builder's document processing models (previously called Form Processing) received a significant accuracy overhaul in Wave 2. Microsoft retrained the underlying models with a larger multilingual dataset, delivering improvements particularly for invoices in non-English languages and for tables with merged cells — two failure modes that frustrated users in Wave 1.

New Prebuilt Models

• Bank Statement model: Extracts transactions, opening/closing balances, and account metadata from standard bank statement PDFs. Supports formats from major US, UK, EU, and Australian banks.
• Purchase Order model: Extracts line items, vendor details, delivery addresses, and PO numbers. Handles both structured and semi-structured PO layouts.
• Identity Document model update: Now handles Indian Aadhaar cards, PAN cards, and UAE Emirates ID in addition to passports and driving licences.

Custom Model Accuracy

Custom document processing models now require a minimum of 5 sample documents (down from 15) to train, and Microsoft introduced an auto-label feature that suggests field bounding boxes based on the document structure. In internal testing, auto-labelled models achieved 85–92% field extraction accuracy after training on 10 documents — a meaningful improvement for organisations processing proprietary document formats.

Desktop Flow Recorder v2

Desktop Flow Recorder v2 is the most significant quality-of-life improvement for RPA practitioners in this wave. The original recorder captured actions at the UI element level with no understanding of intent, producing fragile flows that broke whenever the application changed its layout. Recorder v2 introduces semantic recording — it captures the user's intent (fill this field, click this button, read this value) and generates more resilient selectors.

What Changed

• Smart selectors: Recorder v2 generates multiple fallback selectors per element (by ID, by label, by position, by visual) and Power Automate tries them in order at runtime. A layout change that breaks one selector does not break the flow.
• Table data extraction: Recorder now detects table structures in desktop apps and generates Extract data from table actions automatically, replacing manual row loops.
• SAP GUI recording: Native SAP GUI transaction recording is now stable and generates human-readable action names instead of raw SAP field IDs.
• Image-based fallback removed: The legacy image-based click action (which was unreliable at different screen resolutions) has been retired. Recorder v2 does not generate image-based actions — if an element cannot be reached via UI selector, it surfaces a warning and prompts the developer to handle it manually.

New Connectors and Actions

Wave 2 delivered a meaningful batch of new connector actions across the most-used connectors in enterprise tenants.

SharePoint Connector

• Batch update list items: Update up to 100 list items in a single action without an Apply to Each loop. Dramatically reduces run time for bulk operations.
• Get list item changes: Returns items modified since a given timestamp — eliminates the need to query the full list and filter manually for change-detection patterns.
• Copy file with metadata: Copies a file and its associated metadata columns in one action, resolving a long-standing gap where file copies lost metadata.

Microsoft Teams Connector

• Create meeting with external attendees: Previously, the Teams connector could only create meetings for internal (tenant) users. Wave 2 adds external email support.
• Get meeting attendance report: Retrieves attendance data (join time, leave time, duration) for completed Teams meetings — useful for compliance flows.

SAP Connector Updates

The SAP connector now supports RFC (Remote Function Call) with table parameters, enabling flows to read and write SAP tables directly without a BAPi wrapper. This unblocks several integration patterns — particularly for materials management and finance — that previously required custom SAP middleware.

Process Mining Updates

Process Mining (formerly Process Advisor) received a set of analytical capabilities in Wave 2 that close the gap between Microsoft's offering and specialised process mining tools like Celonis and UiPath Process Mining.

Bottleneck Detection

The new bottleneck detection feature automatically identifies the activities and handovers in a process that contribute most to cycle time. Instead of manually comparing case durations across variants, Process Mining now surfaces a ranked list of bottlenecks with median wait times and the percentage of cases affected. Each bottleneck links to the specific process variant where it occurs most frequently.

Variance Analysis

Variance analysis now lets you compare two process variants side-by-side — for example, cases that completed on time versus cases that were overdue — and highlights which activities, resources, or attributes differ between them. This replaces a previously manual analysis that required exporting data to Power BI.

Direct Integration with Power Automate

Process Mining can now trigger a Power Automate flow when a defined threshold is breached — for example, if the average approval wait time in a procurement process exceeds 48 hours, automatically send a notification to the process owner. This closes the loop between insight and action without requiring manual monitoring.

Enabling Wave 2 Features in Your Tenant

Not all Wave 2 features are on by default. Features in public preview must be explicitly enabled per environment in the Power Platform admin centre. GA features are enabled automatically, but some require admin opt-in before they appear in the flow designer.

Enabling Preview Features

1. Navigate to the Power Platform admin centre (admin.powerplatform.microsoft.com).
2. Select Environments and choose the environment where you want to enable the preview.
3. Go to Settings > Features.
4. Under the "Preview" section, toggle on the features you want — for example, "AI agent flows" or "Natural language approvals".
5. Save. Changes take effect within a few minutes for new flow sessions.

Copilot Availability by Licence

Copilot in cloud flows requires a Power Automate Premium licence (previously Per Flow or Per User with Premium) or a Microsoft 365 Copilot licence. The basic Copilot experience (single-action suggestion) is available on the standard Power Automate licence, but the full multi-step flow generation described above requires Premium or Copilot.

What Comes in Wave 1 2026

Microsoft's release plan for Wave 1 2026 (April – September 2026) previews several capabilities that extend the AI agent story introduced in Wave 2.

AI Agent Flows

The most anticipated Wave 1 2026 feature is AI agent flows — flows that have a Copilot agent as the orchestrator rather than a fixed trigger-action sequence. Instead of a developer defining every step, the agent decides at runtime which actions to take based on the goal and the current state of the data. This is a fundamental shift from deterministic RPA to goal-directed automation.

In the Wave 2 preview available from January 2026, agent flows can be given a goal (e.g., "process all invoices in this SharePoint folder and create matching records in Dataverse") and a set of permitted actions (read files, create records, send email). The agent executes the steps, handles errors, and reports back — without the developer having to define every branch and error path explicitly.

• Natural language approvals: Approvers can respond to approval requests via a Teams message or email in plain English — "approved, please proceed" — rather than clicking a button in the Approvals app. The agent interprets the response and updates the flow state accordingly.
• Cross-tenant flow sharing: Share a flow definition with users in a different Entra tenant. Currently flows cannot be shared outside the owning tenant without packaging them in a solution and exporting manually.
• Expanded Copilot in desktop flows: Copilot assistance for desktop flow authoring — describe the desktop automation task in natural language and Copilot generates the action sequence in Power Automate Desktop, similar to the cloud flow capability in Wave 2.