July was not a month of big showcase demos. Instead, it was a month in which Microsoft 365 Copilot noticeably matured. New models, more powerful notebooks, more context in Outlook, better support in PowerPoint and Excel, and improved agent governance. Prompts and data access clearly demonstrate that Copilot is evolving from an AI assistant into a true work platform.
With Wave 5, Copilot became an autopilot. In July, we will see how this autopilot performs in everyday work. It is less about what AI can theoretically do and much more about how we can work more safely, quickly, and effectively with it every day. In short, Copilot is maturing. Not only is it becoming more intelligent, but also more reliable, more transparent, and closer to the tools we use daily.
The most significant development in July is Microsoft's expanded multi-model strategy. With GPT-5.6 and Claude Opus 5, two powerful models are now integrated into Microsoft 365 Copilot.
Copilot is no longer just a chat window with a model running in the background. Copilot is evolving into a work platform that can utilize the appropriate model depending on the task. Sometimes speed is needed. Sometimes creativity. Sometimes deep reasoning. Sometimes a model that better organizes long documents or complex tasks.
GPT-5.6 is particularly well-suited for broad, everyday tasks, analyses, drafts, and multi-stage projects. In Word, it helps with structuring and developing texts. In Excel, it supports more complex analyses. In PowerPoint, it helps with creating presentation drafts. And in Copilot Cowork, it can manage tasks across multiple tools and files.
Claude Opus 5 complements this approach. Its focus is more strongly on complex knowledge work, long documents, demanding structural work, and multi-stage agent workflows. This is especially exciting for presentations, larger documents, modeled Excel workbooks, and agents in Copilot Studio.
Do not make model selection an expert topic. Create simple guidelines for your organization::
GPT-5.6 for speed, everyday tasks, analysis, and robust designs.
Claude Opus 5 for long documents, complex structural work, demanding presentations, and agent-based workflows.
Auto, if users do not want to decide for themselves.
For important use cases, however, it is worthwhile to deliberately test two models against each other. This way, model diversity does not become more complicated, but more productive.
If I were to show every knowledge worker one feature in July, it would be Copilot Notebooks. Notebooks are AI-powered workspaces. You can collect relevant content, files, notes, and sources there so Copilot can better understand the context of a task. Importantly, Copilot works with the references you consciously add to the notebook. It does not automatically access your entire OneDrive, your emails, or all your Teams chats.
In July, notebooks will become significantly more useful. Word documents, Excel files, and PowerPoint presentations can be created directly from a notebook. Mind maps also help you better visualize topics, patterns, and relationships. This is a major step forward. Collected notes and files will no longer just provide context. They will directly generate a first, usable draft.
You gather a kick-off meeting minutes, a proposal, a few emails, a PowerPoint template, and two reference documents in a notebook. Copilot then uses this information to create a project overview in Word, an initial Excel structure for tasks or risks, and a management deck for the next status call. So you are no longer starting from scratch. You begin with an initial result that you can further develop.
Governance also plays a crucial role here. Notebooks function as clearly defined workspaces. Additionally, generated files can inherit sensitivity labels based on the sources used. These are precisely the kinds of features needed for AI-powered work to scale safely within organizations.
Do not position Notebooks as the new OneNote. Position them as a project hub for collaborative work. Good starting points are:
Outlook is the operational control center for many employees. That is precisely why it is so important that Copilot understands more context within it. Previously, Copilot in Outlook often focused heavily on individual emails or email threads. In July, its scope will be expanded. Copilot will be able to analyze the inbox, calendar, meetings, and other company data that users already have access to. This may sound technical, but it makes a big difference in everyday use. For example, «Summarize this email» will become:
«What are the latest updates on Project X?»
«What open issues have arisen from my emails and meetings?»
«Which appointments next week require preparation?»
«Where do I have free time for focused work?»
This transforms Copilot in Outlook from a mere writing assistant into a true work navigator.
Use this new feature for a small internal enablement session on «Outlook with Copilot.» Three simple prompts are all you need to get started:
«What open issues have arisen from my recent emails and meetings regarding [project]?»
«Summarize the most important updates from my inbox on [topic].»
«Which appointments next week require preparation, and which documents should I review?»
This way, Copilot becomes useful right where many employees begin their workday.
PowerPoint and Excel will also receive important Copilot updates in July. In PowerPoint, Agent Mode with Work IQ becomes significantly more engaging. Copilot can create presentations more effectively based on your work context, including relevant files, meetings, and emails. Existing presentations can also serve as references, allowing for better reuse of content, style, and design. This is particularly valuable for sales, consulting, project management, and leadership communication, as the bottleneck is rarely a single slide. The bottleneck is usually:
Finding the right sources
Understanding the context
Creating a good structure
Maintaining the presentation in line with the corporate design
Clearly conveying the story
Development in Excel continues as well. The stronger integration with Power BI is particularly exciting. Copilot can base analyses on governed Power BI data, including existing permissions. This reduces media breaks and helps avoid unnecessary data exports. At the same time, the new models offer greater flexibility for diverse analytical tasks.
Many of the July updates may be less visible, but they are extremely important for businesses. Agents can be more effectively organized, approved, and controlled. Custom agents can be submitted from the Agent Builder to the Agent Store under «built by your org». They can then be reviewed and approved via the Microsoft 365 Admin Center. This is a crucial step, as agents shouldn't simply be created haphazardly. They require clear processes.
Company-wide prompt publishing in the Copilot Prompt Gallery is another exciting feature. Organizations can create their own prompt collections and make them available tenant-wide. This is vital for adoption, as good prompts shouldn't be created by chance. They should align with real-world workflows, be vetted, and easily accessible. Additional features include MCP-based agents, Federated Copilot Connectors, connector management in the Admin Center, and enhanced capabilities in the Copilot Agent Kit.
Start now with a simple agent governance model. Clarify, for example:
Who is allowed to build agents?
Who reviews agents?
When does an agent go into the store?
Which prompts are officially approved?
Which connectors are allowed?
Who measures usage, quality, and risks?
Another important trend in July: Governance is becoming more visible, not just for administrators, but also for end users. Microsoft is strengthening features related to sensitivity labels, watermarks, and adoption analytics. Generated files can now inherit labels based on their sources. Watermarks may become relevant for AI-generated or AI-modified video and audio content. This is important because AI not only creates content faster, but also introduces new risks:
Unclear content origin
Incorrect distribution
Lack of classification
Uncertainty about sources
Lack of traceability
Labels, watermarks, and clear permissions are therefore not just bureaucracy. They are the foundation for the secure use of AI in organizations. Much is also happening in the adoption area. The Copilot Dashboard offers improved visibility into usage, licensing, and Copilot Chat. For adoption managers, it is becoming increasingly important to view licensed Microsoft 365 Copilot usage and Copilot Chat together.
Do not just communicate governance as an IT issue. Use governance as a signal of trust. Tell users clearly: Copilot can do more. But Copilot operates under your permissions, with labels, with admin control, and with traceable sources. This builds trust. And trust accelerates adoption.
July 2026 points in a clear direction: Copilot is evolving from a spectacular AI assistant into a productive work platform. The key points:
Models are becoming more strategic. GPT 5.6 and Claude Opus 5 bring more possibilities. Companies need clear guidelines for model selection.
Notebooks are becoming project hubs. Documents, spreadsheets, presentations, and mind maps are created from sources, notes, and context.
Outlook is becoming a work navigator. Copilot understands not only individual emails but also more context surrounding the inbox, calendar, and meetings.
PowerPoint and Excel are becoming more production-oriented. Work IQ, Power BI Grounding, inline citations, and model selection make Copilot more powerful for real business deliverables.
Agents need clear governance. Agent Store, Prompt Publishing, Connectors, and Agent Kit demonstrate that the next wave of adoption is also a governance challenge.
Trust is becoming an accelerator. Sensitivity labels, watermarks, and adoption analytics make Copilot safer, more measurable, and easier to explain.