Many AI assistants, one foundation

Matthias Einig 13 November 2026 4 min read

On 12 November 2026, Exchange Online started failing for Microsoft 365 users, first in the Americas and then, Microsoft confirmed, potentially in any region. The disruption lasted more than 16 hours and reached well beyond email.

I want to be transparent about how we read this. Outages happen on every platform. Microsoft’s engineers worked through the night, paused all changes while they investigated, and kept customers updated throughout. This is not a criticism of Microsoft.

It is about something the outage made visible, which I think most AI strategies have not yet accounted for.

The diversification that isn’t

Here is the pattern I see. Organizations rolled out Microsoft 365 Copilot. Then teams asked for Claude, or ChatGPT, or both. Leadership said yes, and it felt like choice. Different assistants for different jobs, and less dependence on any one vendor.

Then each new assistant was connected to the same SharePoint sites, the same mailboxes, and the same Microsoft Entra ID identities. The models are different. The foundation is not.

During the outage, Microsoft 365 Copilot was on Microsoft’s list of affected services. According to reports, the Microsoft 365 connectors that other AI assistants use to reach mail and files were affected too. Organizations running three assistants found they had one dependency.

Every assistant inherits your oversharing. The outage showed it inherits your availability too.

The estate decides what AI is worth

We have said for a long time that you cannot govern AI if you have not first governed what AI accesses. Assistants read at the access you already grant. They answer from the content they find. Their risks are your own data and permission gaps, made instantly visible.

The same is true of value. Gartner found that organizations that govern Microsoft 365 well realize significant Copilot value 26% of the time, against 3% for those that do not. I read that as a statement about the estate, not the model. The model is the same for everyone. The estate underneath it is what differs.

An outage is the same lesson from another angle. For as long as it lasts, everything you have built on the platform shows itself by stopping. During the outage, Teams calls and meetings largely kept working. Teams calendars and scheduling did not, because they depend on Exchange. Most organizations would not have predicted that split, because nobody had mapped it.

Agents change what an outage means

When people lose email, they wait, then pick up where they left off. When an agent loses its platform, it stops halfway through a task. Someone has to find it, work out what it did, and decide whether to run it again.

Gartner expects 150,000+ AI agents in use at the average Fortune 500 company by 2028. I believe that number will arrive earlier. Gartner also reports that 80% of organizations say they lack the governance controls to deploy AI agents at scale.

Put those findings next to the outage. The organizations that recover cleanly from the next one will already know which agents exist, what each one reaches, and who owns it.

What I would ask after any outage

Every outage ends with a post-incident review. Most focus on the service that failed. I would widen it to three questions.

What have we built on Microsoft 365? Every team, site, flow, app, and agent, including the ones built outside IT.
What does each one reach? Which data, which connections, and which AI assistants sit on top of it.
Who owns it today? An active, named owner, not the person who built it and has since left.

If those answers take a week to assemble, the gap is not resilience. It is governance.

Where we sit

Rencore does not keep Microsoft 365 online, and no governance platform can. Microsoft runs the platform. We govern what organizations build and connect on top of it, alongside the Microsoft controls they already use.

Rencore Governance inventories the digital workplace, apps, automations, and AI agents, with Microsoft 365 Copilot as the deepest coverage. Claude Enterprise (preview) and ChatGPT Enterprise (preview) sit in the same view, joined by one Microsoft Entra ID identity. One control plane. Every surface.

Governance is what lets an enterprise say yes to AI. It is also what lets it say yes to more than one assistant without multiplying its blind spots.

Govern the environment. Govern the AI.

See every assistant, agent, and flow on one foundation

Inventory what you have built on Microsoft 365, what each piece reaches, and who owns it.

See your tenant inventory or talk to an expert.

Common questions on this

Does adding Claude or ChatGPT next to Copilot reduce our dependence on Microsoft 365?

Not when every assistant is connected to the same SharePoint sites, mailboxes and Microsoft Entra ID identities. The models differ, but the data, the permissions and the availability they depend on are shared. During the November 2026 outage, Microsoft 365 Copilot was listed as affected, and the Microsoft 365 connectors other assistants use to reach mail and files were reported as affected too. Three assistants on one estate are one dependency, not three.

What should a post-incident review cover after a Microsoft 365 outage?

Most reviews stop at the service that failed. Widen it to three questions. What have we built on Microsoft 365, including the teams, sites, flows, apps and agents built outside IT? What does each one reach, in data, connections and AI assistants? And who owns it today, as an active named person rather than whoever built it and has since left? If those answers take a week to assemble, the gap is governance, not resilience.

Why does a platform outage matter more for AI agents than for people?

A person who loses email waits and picks up where they left off. An agent that loses its platform stops halfway through a task, and someone has to find it, work out what it already did and decide whether to run it again. That is only quick when you already know which agents exist, what each one reaches and who owns it. Gartner reports that 80% of organizations say they lack the governance controls to deploy AI agents at scale.

Do Microsoft Purview and Entra ID already show what we have built on Microsoft 365?

They cover important parts of it. Purview classifies and protects data, and Entra ID governs identities and access. Neither gives one inventory of every team, site, flow, app and agent with a named, current owner, and neither covers assistants such as Claude Enterprise or ChatGPT Enterprise. Rencore Governance adds that inventory alongside the Microsoft controls you already use, joining Microsoft 365 and those assistants through one Entra ID identity per person.

Can a governance platform keep Microsoft 365 online during an outage?

No. Microsoft runs the platform and only Microsoft can restore it. What governance changes is how quickly you understand the impact and recover. Rencore governs what organizations build and connect on top of Microsoft 365, so after an outage you can see which agents, flows and apps depend on the affected services, what they reach and who to call, instead of assembling that picture by hand while the incident is still open.

Last updated 13 November 2026

About the author

Matthias Einig

Matthias is a former 10-year Microsoft MVP, co-founder, and CEO at Rencore. His mission is to help enterprises stay in control of their Microsoft collaboration technology. Matthias and his team achieve this by providing insights, advice, and actionable ways to manage platform growth. Matthias is speaking at many conferences and community events all over the world and is one of the organizers of the European Collaboration Summit.