Copilot RPA: Researcher, Opal & Studio

Key insights

• AI RPA and Computer Use: AI agents now interact with apps like a person — they look at the screen, reason, click, type and navigate instead of relying on fragile recorded clicks.
  They replace brittle scripts with UI-aware automation that handles multi-step tasks across systems without APIs.
  
• Researcher with Computer Use: designed for deep, multistep research that gathers, analyzes and cites sources while operating in a secure virtual environment.
  The agent can use visual and text browsers plus a terminal, perform sign-ins with user confirmation, and support additional models (e.g., Anthropic) when enabled.
  
• Project Opal: a personal background robot that runs asynchronously on an Entra-joined Windows 365 Cloud PC, working across browser apps while you keep the option to “Take Control” and stop or resume it at any time.
  It’s aimed at individual, hands-off automation of routine web tasks.
  
• Copilot Studio (maker/IT surface): generally available (May 13, 2026) for commercial Power Platform geographies and built for enterprise automation and governance.
  It includes Azure Key Vault for credentials, Purview audit logging, DLP, website allow-lists and human-in-the-loop approval checkpoints for secure deployments.
  
• Cowork Edge browsing (preview): a lightweight, in-flow option that runs in a hidden Microsoft Edge tab and reuses your single sign-on and existing sessions.
  The agent has exactly the same site access you have, making short web tasks fast while preserving session boundaries.
  
• Benefits and best-fit scenarios: this approach executes real workflows where APIs don’t exist, improves coverage for complex tasks, and reduces fragility from pixel-based scripts.
  Choose the surface by need: Researcher for deep analysis, Opal for personal background work, Copilot Studio for enterprise-grade automations, and Cowork Edge for quick in-context web tasks — all under governance and approval controls.
  

Video Report: Microsoft’s AI RPA Push Explained by Szymon Bochniak (365 atWork)

Szymon Bochniak of 365 atWork lays out Microsoft’s emerging approach to robotic process automation in a recent YouTube video, and our newsroom summarizes the key points for readers. The video maps how Microsoft is moving from brittle, click-recording scripts to intelligence-driven agents that can see and act on the screen like a human. Consequently, this shift changes who can automate work and how automation is governed across enterprise environments.

Bochniak compares legacy RPA with the new model, and then walks viewers through four product surfaces where Microsoft is deploying this capability. He also flags preview and general availability statuses, and highlights names such as Project Opal, Copilot Studio, Cowork Edge browsing, and the research-focused Researcher. Importantly, Bochniak notes that some names may be informal and that product positioning can evolve as Microsoft refines the features.

What is AI RPA and computer use?

Bochniak explains that classic RPA depended on recorded clicks and screen coordinates, which often broke when interfaces changed. In contrast, AI RPA or computer use uses an agent that interprets the visual interface, reasons about the task, and performs actions like a person would.

This agent-based approach allows automation to act where no API exists, bridging web pages, desktop apps, and cloud services. As a result, Copilot moves beyond answering questions and toward executing real, multistep workflows on behalf of users while preserving an audit trail and governance controls in many scenarios.

Four Copilot Surfaces: Who they serve and how they differ

First, Bochniak covers Project Opal, which he describes as a personal background robot that runs tasks asynchronously on an Entra-joined Windows 365 Cloud PC. Users can ask Opal to perform jobs while retaining the option to “Take Control” through remote desktop, ensuring they can intervene if needed.

Second, Copilot Studio targets makers and IT teams; it became generally available on May 13, 2026, across commercial Power Platform regions and includes enterprise controls such as Azure Key Vault credentials, Purview audit logging, DLP protections, website allow-lists, and human-in-the-loop checkpoints. Third, Cowork Edge browsing offers a lightweight, in-flow option in a hidden Microsoft Edge tab that reuses the user’s single sign-on, so the agent has the same access as the user and no extra privileges. Finally, Bochniak examines Researcher and a Researcher variant with computer use, which are designed for deep, source-cited research and can execute UI-driven steps in a secure virtual environment when needed.

Tradeoffs: Security, governance, and usability

Bochniak’s coverage highlights important tradeoffs when organizations adopt AI-driven computer use. On one hand, the capability expands automation to tools lacking APIs and reduces brittleness compared with classic RPA, which improves productivity; on the other hand, it raises security and privacy questions because agents may interact with sensitive sites or credentials.

Microsoft addresses some concerns in Copilot Studio with enterprise controls like credential vaulting and audit logging, and by placing long-running agents on isolated Cloud PCs for Project Opal. Nevertheless, businesses must balance tighter governance against the agility that lightweight options like Cowork offer, since stricter controls can slow deployments and add operational overhead.

Challenges and operational considerations

Bochniak points out practical problems organizations will face when rolling out computer use. First, UI changes still present a reliability risk, so teams need robust monitoring and recovery strategies. Second, authentication flows and multi-factor prompts require careful handling to prevent failures or exposure of secrets.

Moreover, human-in-the-loop checkpoints reduce risk but increase manual steps, which can limit full automation in high-volume scenarios. Finally, testing agents across diverse environments and keeping audit trails that meet compliance needs remain nontrivial tasks that IT and security teams must plan for.

Implications and next steps for organizations

According to Bochniak’s video, organizations should pilot these features with clear governance rules and limited scopes to gather lessons before wider rollout. Furthermore, involving security, compliance, and business owners early helps strike a pragmatic balance between automation speed and risk control.

Overall, Microsoft’s move toward agent-based computer use in Copilot promises to widen automation possibilities while forcing new conversations about access, auditability, and resilience. As these surfaces mature, businesses will need to weigh the benefits of reduced manual work against the operational costs of safe, reliable deployments.

Keywords

AI RPA Copilot, Copilot Studio automation, Researcher Copilot workflows, Opal AI assistant, Copilot Cowork integration, RPA for knowledge work, Microsoft Copilot productivity, AI-powered research automation