How Microsoft’s New Copilot Can “Use a Computer”—and What That Means for Healthcare

Two Decades of RPA Expertise, Reimagined for the Agentic Era

Apr 17, 2025

Executive Summary

After 30 years of coding through more hospital IT upgrades than I care to admit, I thought I’d seen it all—until AI agents started clicking buttons faster than I can find my reading glasses. True Story.

Microsoft's recent launch of the "computer use" feature in Copilot Studio introduces a groundbreaking form of automation: copilots that can see and interact with user interfaces like humans. This interface-level automation mirrors OpenAI's Operator platform, which enables agents to operate computers using screenshots and reasoning.

As explored in Section, School's compelling article on "bounded" versus "open" agents, Microsoft and OpenAI offer two complementary visions for AI-driven work automation. Healthcare executives stand to benefit enormously from these innovations, particularly in administrative and compliance workflows that rely on legacy systems and manual tasks.

This article explores the significance of this shift, compares Microsoft's bounded Copilot Studio agents with OpenAI's open-ended Operator agents, and offers healthcare-specific use cases that demonstrate the transformational potential of agentic automation.

Microsoft’s Secret Sauce - From Macros to RPA to Agentic AI

Let’s set the record straight, Microsoft’s journey into automation didn’t start yesterday. It began with the humble Windows Recorder and Office Macros—those early tools that let us automate the boring stuff back when dial-up was king and “cloud” meant the weather. These macros laid the foundation for automating repetitive tasks, giving users a taste of what was possible when you let the computer do the heavy lifting.

Fast-forward to the era of Robotic Process Automation (RPA): Microsoft took automation to the next level with Power Automate. By integrating RPA—especially after acquiring Softomotive and launching Power Automate Desktop—Microsoft made it possible to automate complex, cross-application workflows, even in legacy environments where APIs were just wishful thinking. Power Automate’s drag-and-drop designer, embedded recorders, and hundreds of connectors democratized automation for everyone from spreadsheet warriors to enterprise architects.

Now, Microsoft is marrying this mature RPA foundation with OpenAI’s Operator-style agentic intelligence. This isn’t just another incremental upgrade—it’s a tectonic shift. In my professional opinion (and after 30 years of living in the trenches of hospital IT), Microsoft’s new AI agents are standing on the shoulders of decades of automation know-how. They can see, reason, and act on any interface, combining the reliability of RPA with the adaptability of modern AI. That’s a competitive edge you can’t fake.

1. Interface-Level AI Agents: A New Era of Automation

Gone are the days when automating a workflow meant begging your vendor for API access, only to be told it’s coming “next quarter”—for the fifth year running. Microsoft’s new Copilot Studio “computer use” capability and OpenAI’s Operator agent break this limitation. They allow AI agents to see the screen, understand the interface, and act on it—clicking buttons, entering data, navigating tabs—without needing any backend integration.

This is a radical shift. These systems can mimic the way a person uses a computer, enabling automation across outdated portals, legacy EHRs, or third-party vendor software with no API access. Importantly, these capabilities keep a human in the loop, allowing for oversight in clinical or high-stakes administrative contexts.

Having spent decades untangling spaghetti code and legacy databases, I can confirm: if an AI agent can survive a day in a hospital’s EHR, it deserves a white coat.

2. Two Paths to Agency: Microsoft vs. OpenAI

Section School’s article "OpenAI Operator vs Microsoft Copilot Studio" smartly frames the difference as a spectrum between bounded and open agents:

Microsoft Copilot Studio introduces a bounded agent. Think of it like your trusty for-loop—predictable and reliable. This model operates within well-defined workflows, governance rules, and user-triggered events. It is highly predictable, designed for enterprise environments where compliance, reliability, and auditability are non-negotiable.
OpenAI’s Operator represents an open agent. More like a recursive function: powerful, but handle with care (and maybe a stack overflow warning). It is trained using reinforcement learning to operate more autonomously, with long-horizon planning and adaptability to unfamiliar interfaces. It thrives in exploratory or complex workflows, where tasks evolve and rules aren’t rigid.

Both are valuable—but the right tool depends on context. For healthcare, where regulation and risk dominate, bounded agents may be ideal for most use cases. However, open agents can offer powerful advantages in research, population health analytics, and dynamic patient journey modeling.

3. Real Healthcare Use Cases

Revenue Cycle Automation: Agents can log into payer portals, retrieve remittance information, check claim statuses, or submit appeals—all tasks that today require costly FTEs or manual workflows. Revenue cycle teams can finally retire their sticky-note armies and let AI handle the click-fest of payer portals. Bonus: fewer carpal tunnel claims.

Compliance and Attestation: For NIH or CMS reporting, agents can interact with fragmented HR, finance, and scheduling systems to gather and format data for attestations or audits.

Referral and Intake Coordination: Agents can act as digital assistants, retrieving updates from EHR portals, uploading referral packets, or scheduling follow-ups when backend access is restricted.

Admin Analytics Dashboards: Agents can unify data from multiple siloed systems and screen-based apps to populate real-time dashboards for department heads or operational leaders.

These are just a few places where agentic AI can tag in as your digital colleague—think less ‘robot overlord,’ more ‘helpful intern who never takes a lunch break or forgets their HIPAA training.’

4. Strategic Implications for Executives

For healthcare leaders, the key takeaway is that AI is no longer confined to chat or static analytics. With Copilot Studio and Operator-style agents, we now have AI that can see, think, and act.

Start with bounded agents: Use Copilot Studio for controlled, low-risk automation across common administrative workflows.
Explore open agents for innovation: Pilot Operator-like tools in R&D or population health, where adaptability is essential.
Bridge the integration gap: These agents reduce the need for expensive API development or vendor coordination.
Keep governance top of mind: Adopt strong AI governance and human-in-the-loop models to ensure safety, compliance, and trust.

Ever wish your compliance audits could run themselves while you grab a coffee? With agentic AI, that’s not just a caffeine-fueled fantasy.

Conclusion: The Future is Agentic—and It's Already Here

This isn't speculative technology—it's in preview and use today. For hospital systems still running on software old enough to remember Y2K, these AI agents could be the biggest breakthrough since someone said, ‘Hey, what if we put patient charts on a computer?’ Please consider subscribing to my free Substack.

Citations:

Research from Perplexity, ChatGPT, and Grammarly

How Microsoft’s New Copilot Can “Use a Computer”—and What That Means for Healthcare

Two Decades of RPA Expertise, Reimagined for the Agentic Era

Executive Summary

Microsoft’s Secret Sauce - From Macros to RPA to Agentic AI

Citations:

Discussion about this post