The Invisible Work Behind Every Tool You Use: The M365 Outage

Microsoft 365 went down, and Copilot went with it

A Microsoft 365 outage that began Thursday took down Outlook, Teams, SharePoint, and Copilot across the Americas, and problems reached Australia by Friday morning.

BleepingComputer reported Microsoft's wider advisory started at 1:26 p.m. Eastern, after mailbox problems earlier in the day.

Microsoft said a core process that sets up secure connections became overwhelmed, and servers began failing their automated health checks. The company said there is no indication of a cyber attack. As of early Friday, Microsoft still listed the incident as a service degradation, with recovery partial and uneven.

Most people never think about that process at all, and that is the point. Email just opens. The meeting just starts. The assistant just answers. Behind each of those moments sits a chain of systems: capacity planned months ahead, health checks running through the night. People watch those systems around the clock so the rest of us never have to think about them. That work is invisible on the days it works, which is nearly every day, so it gets taken for granted.

The damage reached past Microsoft's own products. Reuters reported the outage also broke connectors that OpenAI and Anthropic use to reach Microsoft 365 data. When the platform underneath goes down, the AI tools built on top of it stop too, whoever sells them. Copilot did not fail on its own. It inherited the outage from the data layer it reads.

That is the adoption question nobody puts in the pilot plan. Teams now route real work through these tools: email triage, meeting notes, and agent runs against live mailboxes. When the platform stops, that work stops with it, and the measure that matters is hours of work delayed. The practical check is a fallback path people actually know, plus agent jobs that can pause and resume instead of failing quietly. Thursday's outage is the argument for writing that down before the rollout scales.

Jensen Huang resigned, and Colette Kress holds the company

Yesterday's brief left NVIDIA's board weighing Jensen Huang's future. On Thursday, Huang resigned, ending a 33-year run at the company he co-founded. Reuters reported the announcement came about 42 minutes before markets opened. Chief Financial Officer Colette Kress takes over as interim chief executive while keeping her finance job, and the board named Harvey Jones, a director, as its first-ever chair.

The board had opened an independent investigation into reports that Huang was the target of a foreign intelligence effort, and NVIDIA gave no reason for the resignation in its announcement. The company said it expects operations and its product roadmap to continue without interruption. Shares fell more than 10% before the open and closed the day down 4.3%. Huang leaves the board too, keeps a stake of about 3.61%, and has agreed to testify before Congress before Thanksgiving.

Kress has run NVIDIA's finances since 2013, and Reuters reported she was known as the executive most willing to push back on Huang's biggest bets. Her first public test comes Tuesday, November 17, when NVIDIA reports quarterly results as scheduled. President Trump also posted that Tim Cook should replace Huang. Cook has not responded, and the board's search for a permanent chief executive is already under way.

Anthropic will let companies run Claude's agent service on their own servers

Anthropic opened a private beta on Thursday that lets companies run the whole Claude Managed Agents service inside their own cloud or data center. Anthropic's announcement covers the agent harness, the sandboxes, the credentials agents use, and the full session history.

One boundary stays where it was. The model itself does not move. Each model call still goes to a provider the company picks, and Anthropic says it has no remote access to a customer's self-hosted server or its logs. The limits are real too. Access runs only through Anthropic's account team, for eligible organizations, with no pricing published and no date for general availability. Some hosted features are missing, and nothing migrates from the hosted service, so agents, credentials, and session history all have to be rebuilt.

That is the standing objection to agents in regulated work: where the data sits, and who can see it. The question to ask any agent vendor is now concrete: which parts run inside our walls, which parts call out, and who holds the keys either way?

Also today, Google launched Gemini 4 Hydrogen, a fast model for coding and everyday work, priced at a flat $1.50 per million input tokens. Google put it into Search on day one. OpenAI opened two security products under the name Daybreak, including a red team service in early access. Meta's Muse agent began placing phone calls to US businesses Thursday, after testing with a small group since September. Paid plans start at $20 a month. Tell it who to call: the cable company, the DMV, a restaurant that takes no bookings. Grok Bot released the same feature earlier this month. It dials a US business and holds the conversation. A live transcript runs during the call, with a recap after. One honest limit: some businesses hang up when they realize the caller is an AI. Agents are starting to do their work through the same channels people use, including the phone.

The work that failed on Thursday is the work nobody writes about on the good days. The systems held for years because people planned the capacity, watched the checks, and fixed the small problems before they became Thursday. That is the standard worth copying: know what your team depends on, know who is watching it, and know the fallback before the day it stops. NVIDIA's week makes the same point from the other side. A bench that can act in a day is just another kind of fallback, built before it was needed.

About this brief: This brief was researched and drafted by my personal AI assistant and approved by me before publishing. Product details change fast; check primary sources before you build on them.

Want more with the Tech Simplified crew? Subscribe on Substack.

Get the week's briefs in one email. The Briefcase packs Monday to Friday's AI Briefs into one Sunday email.

Follow me on LinkedIn for project updates, ideas for using AI, and lessons you can bring to your own work.