Copilot Wave 3: What the New Wave of Agents Means for Businesses

AI Agents in Microsoft 365: What Changes with the New Copilot Capabilities

The latest Copilot updates revolve around a fundamental shift: AI is no longer meant to simply respond to prompts, it is expected to act independently. AI agents are directly integrated into Office applications and take over recurring tasks in the background. Through Microsoft Work IQ, they can access the user’s full work context, including emails, documents, meetings, and Teams messages. This enables automated status updates, structured summaries, and pre-sorted inboxes without requiring users to initiate every step manually.

With Copilot Cowork, Microsoft is going a step further. Users can delegate entire tasks: Copilot breaks them down into subtasks, coordinates the required applications, and delivers a completed result. The process remains transparent and can be adjusted or stopped at any time. Tasks can also run in the background over extended periods. Copilot Cowork is currently available in preview as part of the Microsoft Frontier program.

Governance for AI Agents: Security, Compliance, and Clear Accountability

In addition to Microsoft-provided agents, organizations can build their own agents using Copilot Studio or integrate third-party solutions via connectors. To manage this growing agentic workforce, Microsoft is introducing the Microsoft Agent 365 platform. For the first time, AI agents are treated similarly to users: they are assigned identities, access rights, and licenses. The platform is integrated with Microsoft Entra, Microsoft Purview, and Microsoft Defender, enabling administrators to gain real-time visibility and enforce policies centrally.

Governance is critical. As soon as agents are deployed across departments or organization-wide, the same requirements apply as for any mission-critical application: risk assessment, clearly defined ownership, and integration into ticketing systems. Underestimating this can result in uncontrolled AI activity with corresponding compliance implications.

AI Agents in Practice: Typical Challenges

In practice, AI agents rarely fail because of the technology itself. The most common causes of poor outcomes are an unclear scope, imprecise instructions, and insufficient data availability. An agent is not a generalist but a specialist designed for a specific task and a clearly defined data set. Content that sits behind access barriers or within expandable elements in SharePoint is simply not accessible to them.

Operational overhead is another factor. Every agent requires an owner responsible for maintaining its function and data foundation, as well as monitoring model updates. Without clear accountability, agents quickly become unmanageable. In addition, not every use case justifies a fully autonomous agent. In many cases, a lean automation approach with targeted AI components is sufficient. Organizations that do not assess this upfront often invest more than necessary.

Model Selection and Licensing: More Options, Greater Complexity

With the latest Copilot capabilities, Microsoft is opening its ecosystem to models from other providers. In addition to OpenAI’s GPT models, customers can now select Claude models from Anthropic within Copilot Chat, Copilot Studio, and certain agents. No single model performs equally well across all scenarios, and the market is evolving rapidly. Organizations that want to make informed, use-case-specific choices need to continuously monitor and evaluate developments.

At the same time, licensing is becoming more complex. Between integrated per-user licenses, pay-as-you-go models, and the new Microsoft 365 E7 suite priced at USD 99 per user per month, the most cost-effective option is rarely obvious. The E7 suite bundles Microsoft 365 Copilot Premium, Agent 365, Entra, and E5 security capabilities into a single license. Whether it is financially viable depends on the specific usage profile and use cases.

Adopting AI Agents: How SoftwareOne Supports Organizations

The new Copilot capabilities require strategic decisions: Which model fits which use case? Which licensing approach is economically viable? How can agents be operated securely and in compliance with regulations?

As one of the first partners with the Microsoft Frontier Partner Badge and a Microsoft Copilot specialization, SoftwareOne helps organizations answer exactly these questions. The team includes more than 1,000 Microsoft specialists and 18 Microsoft MVPs, complemented by over 30 years of licensing expertise.

Modular advisory services are designed to support organizations at every stage, from initial compliance checks and Copilot Chat QuickStart workshops to readiness assessments, pilot projects, and ongoing support as a trusted advisor. SoftwareOne continuously monitors market developments, filters relevant updates, and prepares them for practical application, enabling organizations to focus on their core business.

FAQ: Common Questions from Practice

What exactly are AI agents in Microsoft 365 Copilot?

AI agents in Microsoft 365 Copilot are specialized digital assistants that can carry out tasks independently. They access work context such as emails, documents, and meetings, and automate multi-step workflows directly within applications like Outlook, Teams, or Excel.

What prerequisites should organizations meet to use Copilot agents?

Organizations need a structured data foundation, clearly defined access rights, and robust governance and security frameworks. Since AI agents operate on company data and act independently, compliance, identity management, and monitoring must be addressed from the outset.

In which scenarios does the use of AI agents in business environments make sense?

AI agents are most effective in recurring, clearly defined processes with a stable data foundation. Not every use case requires a fully autonomous agent; in many situations, targeted automation with AI components is sufficient.