Microsoft 365 went down for more than 24 hours, and the first thing Microsoft told customers was what it *wasn't*. In a post on X, the company said: "This was a technical issue not related to cyber or AI/agents." The cause, per Cointelegraph's report, was a problem introduced during an earlier network capacity upgrade. The take for builders: the outage was a plain infrastructure failure, and it still reached every agent that depended on that infrastructure.
What happened
Per Cointelegraph, the disruption began on a Thursday and hit Exchange Online mailboxes, plus features tied to Outlook, Teams and Microsoft Copilot. Government agencies in Japan, Singapore and the United States reported email problems. Microsoft said it fixed the underlying issue on Friday and restored service gradually, checking stability before each next stage. On Saturday it said it had "restored service to the remaining impacted customers and regions" and pointed customers to incident MO1550821 in the Microsoft 365 admin center.
Microsoft apologized and said it will publish a full post-incident report once its review is complete. That report has not been cited here because, as of the source article, it had not been published. Until it lands, "a network change went wrong" is Microsoft's own account, not a verified root-cause analysis.
The part that matters for agents
The article also notes that connections between Microsoft 365 and AI assistants were affected. Afterward, OpenAI said the Microsoft 365 connector in ChatGPT was operating normally again, and Anthropic marked its Microsoft 365 connector and its Claude for Outlook incident as resolved.
Read that carefully. Microsoft says the cause had nothing to do with AI. The agent-side effect was downstream: a connector is only as available as the service behind it. If your agent reads mail, drafts replies or triggers actions through a hosted API, an upstream outage is your outage.
Nobody needs to panic about this, and nobody should shrug it off either. "Agents" tend to get blamed for new kinds of failure. This one is a reminder that the old kinds still work fine.
Why this lands differently onchain
Onchain agents have a version of this problem that is easy to forget because the chain itself is the famously resilient part. The settlement layer keeps producing blocks. Everything around it does not:
- ▸RPC endpoints. Your agent talks to the chain through someone's node.
- ▸Data and price feeds. Hosted services your agent treats as ground truth.
- ▸Identity and notification channels. Email, chat and messaging connectors that carry approvals or alerts.
- ▸Model providers. The brain is usually an API call away from its own outage.
A wallet with spend limits does not help if the approval email never arrives. A clever strategy does not help if the feed it reads is dark. The prompt is the cute part; the stack is where the work lives, and the stack includes every hosted dependency you did not write.
What a builder can do with this
None of this is exotic. It is checklist work:
1. List every hosted dependency your agent touches, including the boring ones like mail and chat connectors. 2. Decide the failure mode for each. Pause, retry, fall back, or ask a human. "Keep going with stale data" should be a choice, not an accident. 3. Separate read paths from write paths. An agent that cannot read should not be guessing its way into moving funds. 4. Keep a second route for anything on the critical path, such as a backup RPC or an alternate notification channel. 5. Log the outage, not just the error. When a dependency is down, your agent's history should say so.
This is the unglamorous side of "build any crypto agent." Sato maps the stack, the skills, MCPs, wallets, payment rails and frameworks, so you can see what your agent leans on before an incident shows you. The Sato Score is a transparency and liveness signal for a listing. It is not a safety or quality grade, and it will not tell you whether a vendor's next network change goes well.
What to watch
- ▸Microsoft's post-incident report. Microsoft says it is coming. It should say more about the network capacity upgrade and what changed, and it is the document to read instead of the headline.
- ▸How connector providers describe their own status. OpenAI and Anthropic each posted resolution notes for their Microsoft 365 connectors. Watch whether agent platforms start separating "our service" from "the service we connect to" on status pages.
- ▸Your own dependency list. If an outage like this had hit last Thursday, would your agent have paused cleanly or kept acting on stale inputs?
Start with the stack. Describe the agent you want and map what it depends on at satohub.ai/build.