Cloud AI has bad days. Pick a second assistant from a different company, load it with your context, and know where to check status.

Your AI assistant is going to go down, so set up the spare before it does. On 3 September, Anthropic confirmed elevated errors across multiple Claude models. Just over an hour later, OpenAI confirmed a major ChatGPT outage across more than a dozen components, from login to file uploads to voice mode. Two of the biggest assistants, one bad day. In July, ChatGPT went down worldwide for roughly 50 minutes and wouldn't even load old conversations.

That's the deal when your thinking partner lives in somebody else's data center. Some mornings the whole thing hiccups, and anyone who built their workday around one chat window just sits there hitting refresh.

Pilots solved this a long time ago. They pick an alternate airport before the plane leaves the gate, not while it's circling in fog. Your AI setup deserves the same courtesy, and it takes about ten minutes.

Start by picking a second assistant from a different company. If you live in Claude, make ChatGPT or Gemini your alternate, and the other way around. The 3 September pileup is why it has to be a different vendor — a backup that goes down with your main tool isn't a backup.

Then load it before you need it. That's the context file from our Lesson on keeping your context portable — who you are, how you work, the standing facts about your projects. Paste it into the second assistant's custom instructions or a project today. Switching then takes a minute instead of a morning spent re-explaining yourself to a stranger while your deadline ticks.

Last, bookmark the status pages: OpenAI, Anthropic and Google Workspace, which covers Gemini. They tell you in seconds whether the problem is your Wi-Fi or their servers. When it's them, stop retrying and switch.

The part most people skip is the fire drill. Do one real task in your alternate this week, while nothing's on fire. You'll find the lines missing from your context file, the prompt habits that don't carry over, and the one thing the backup actually does better. Better to learn that on a quiet Tuesday than during a client deadline.

A spare earns its keep on good days too. Models go easy on their own work: a study across seven model families found the stronger the model, the more it favors its own answer when that answer is wrong. In an October test of 900 review sessions with planted errors, a same-model review plus a cross-model review caught 57% of the mistakes, against 43% for two passes by the same model. Before anything important goes out, paste it into the spare and ask what the first one missed.

And if everything's down at once? Your context file still works as a checklist for doing the job by hand, and yesterday's Lesson covered which tasks you can run on your own machine.

Test your spare before Friday, then hit reply and tell me which assistant you picked as your alternate.