Table of Contents
Many Microsoft Learn GitHub Repositories to Become Private
I am disappointed, though not surprised, at the September 23 announcement about changes Microsoft is making to the ability for external people to contribute to Microsoft Learn documentation via GitHub. Some useful commentary on the announcement comes from John Flores, who was a documentation writer (aka “Senior Knowledge Engineer”) for Entra ID from February 2017 to July 2026.
Entra ID’s documentation has improved dramatically over the last decade. I think a lot of the improvement came from internal processes within Microsoft allied to a very active and open technical community who dissect new features and functionality to analyze and understand exactly how the software works. Some of that feedback ends up in external blogs, and some used to flow into Microsoft through GitHub requests. Documentation based on hard experience is the best sort, and documentation honed by feedback from real experts in the field is gold dust.
Microsoft plans to retire many public documentation repositories by the end of 2026, moving them from public visibility and contribution to private management. Microsoft plans to retire many public documentation repositories by the end of 2026, moving them from public visibility and contribution to private management. Sample code repositories are unaffected, as is anything associated with open source projects. Microsoft says the goal is to maintain trusted, high-quality content and establish clearer criteria for public repositories. However, they do not explain why private repositories are better suited to achieving that goal.
When a repository is retired from public view, people outside Microsoft won’t be able to access the source documentation files and suggest changes. People won’t be able to see what issues the technical community reports, how Microsoft responds, or how discussions evolve over time. For example, external contributors wouldn’t have been able to discuss the ongoing assembly clashes issue involving the Microsoft Graph PowerShell SDK with the developers (a fix appears to be imminent).
External contributors also won’t be able to track the changes Microsoft makes to repositories when new features are being added. There are quite a few utilities that monitor GitHub for changes posted to documentation repositories to highlight new information. These utilities will cease working after a repository becomes private.
Answering Difficult Questions
As John Flores points out, the hardest documentation problems are how to answer questions that customers face in the real world. He says that the answers came from experienced writers partnering with engineering teams to translate product knowledge into something that tenant administrators could understand. Community contributions helped to close the gaps between what Microsoft believes software should do and what actually happens when software is exposed to the hard reality of real-world operations.
For example, the Microsoft Learn page covering the use of Entra ID groups to manage role assignments contains a note about the old Exchange admin center that disappeared from Microsoft 365 in 2022. Anyone with a GitHub account can change the text to remove the old text, which is what I did (Figure 1). It’s a good example of a detail that a content maintainer might not be aware of, which is why outdated content can find its way into documentation (here’s an example).
I’m sure that Microsoft management has a plan to offset the loss of community contributions. It’s probable that the plan involves greater use of AI to generate and maintain product documentation, which may help explain why Microsoft’s documentation community has shrunk so dramatically over recent years. AI can generate basic product documentation, but it cannot replace the insight of experienced writers who have worked alongside engineers for years. AI just doesn’t know the questions to ask.
AI and Community Contributions
But one thing that AI is very good at is finding and analyzing community contributions published on blogs like Office365ITPros.com. AI has always excelled at ripping off the accumulated knowledge of the planet by searching the internet for information. The problem is that AI often cannot distinguish between old and new knowledge, or accurate and flawed knowledge that it finds on a blog or other website.
If Microsoft management thinks that they can do away with community contributions that are parsed and examined through the pull request mechanism and replace it with information scoured from random blogs, I fear for the quality of the outcome. Removing the humans from the process will certainly eliminate some cost, but the potential impact on quality of Microsoft Learn documentation is unpredictable.
Microsoft Learn Quality Set to Fall
I cannot see quality improving, unless Microsoft’s tools are capable of testing every assertion and every code example fetched from external blogs before appearing in documentation. Given Copilot’s well-documented tendency to generate PowerShell code containing errors, non-existent cmdlets, incorrect parameters, and similar problems, I’m at a loss to understand how Microsoft Learn could do any better.
Perhaps Microsoft has concluded that the benefits of openness no longer justify the operational costs. If so, I believe that decision underestimates the value the community brings to documentation quality. I also see little evidence that issues raised in alternative forums, like Microsoft Q&A and the Microsoft Technical Community, will be any more successful in improving documentation.
Software moves so quickly in the cloud that perhaps striving for accurate, precise, and informative information about how software really works is an impossible dream. The Microsoft 365 for IT Pros team has documented our understanding of Microsoft 365 since 2015. Our experience has consistently shown that the best documentation emerges from a combination of engineering knowledge, professional writers, and an engaged community. Removing one of those elements is unlikely to improve the result.
The nice thing about generative AI tools is that they can generate information based on what’s gone before. The bad thing about generative AI tools is that they can’t create new thinking or insights about how technology works (or doesn’t). When we write the Microsoft 365 for IT Pros eBooks, we depend on hundreds of years of real-world experience, knowledge, and intuition to analyze and explain how Microsoft 365 really works. If you understand the basic principles about Entra ID, Exchange Online, SharePoint Online, Teams, the Microsoft Graph, and more, you’ll be able to figure out the value of new features as Microsoft adds them to the platform. All for less than ten copies of black coffee.