Copilot went down Friday. Nobody knew what it touched. 2,600 complaints on Downdetector in under an hour. Blank responses. Failed prompts. No fallback. Microsoft spent three years building the dependency. Nobody spent three years mapping what it touched. A misconfigured load-shedding algorithm in the Azure OpenAI backend silently rerouted traffic into a wall. West US 2. Degraded across the region. The technical failure is not the story. The story is that 62% of enterprise Copilot users call it critical to their daily work, and almost none of them could answer a basic question Friday morning: What workflows break when this goes dark? Which teams stop. Which decisions stall. Which downstream systems were quietly depending on a prompt completing. That map did not exist. Embedding is not integration. Microsoft distributed Copilot across Windows, Teams, Word, Excel, Outlook, and Edge. Distribution is not a dependency graph. Criticality is discovered, not declared. The Friday outage was a free audit. Most teams did not run it. Fallback is a topology question. You cannot design resilience for a system you have not mapped. Microsoft will fix the algorithm by Monday. The map still won't exist Tuesday. If your AI stack went dark for three hours tomorrow, could you name the ten workflows that would stop first?
Distribution is not integration. That line should be in every onboarding deck.
We had nothing.
I have zero visibility into what my stack actually touches day to day.
Ran the dependency audit after reading this. Results were not good.
This is on IT teams, not Microsoft. You chose to depend on it.
The closing question hit.
The trust problem outlasts the tech problem every single time.
The free audit nobody asked for.
No map, no fallback. That's the real outage.
The outage didn't create the problem. It just made it visible. The dependency existed before Friday. The map didn't. That's the real audit. Not what broke. What was never documented in the first place.