When ChatGPT Is Down, Your One-Person Company Shouldn't Be: Add a 20-Minute Fallback

If your one-person company leans on a chatbot for drafting, support, or internal ops, the scary moment is the quiet hour you spend waiting, re-prompting, or deciding what to do manually. That hour is where the margin disappears: the client message stays open, the small decision grows, and the day starts to feel improvised. The outage is the event; the hour is the real expense.

Multiple major AI chatbots suffered a service outage that affected users; Down Detector reported problems in four popular chatbots; OpenAI's status page reported severe errors in ChatGPT and Codex, while Claude's outage affected several named Claude models; ChatGPT was reported as the first service to return to normal operation.

The lesson is to treat AI like a utility: something you use, without emotional dependence. You do not plan your entire day around the lights being perfect. You keep a battery, a backup plan, and a manual switch. The same logic applies to AI in a solo business. If a model is your drafting assistant, your support triage, or your content engine, you need a way to keep moving when it is unavailable. First principles are simple: if the tool stops, the work still needs a path.

The goal is to make the fallback smaller than the outage. You want a system that takes minutes to activate, not a new project to build after the fact. It is a boring, repeatable habit: a saved prompt, another option, and a short manual path that lets you finish the minimum useful work. It is the mental tax of re-entering the work, the client message that sits unanswered, and the small decision that becomes a large one because you waited too long. The margin you protect is the ability to stay in control of your day.

Build the Outage Fallback Card

The Outage Fallback Card is a single page you can find in a hurry. It should be a card, not a strategy document. It should be a card with the critical task, the versioned prompt, the fallback model, the manual checklist, and the time budget. Keep it in the same place you keep your invoice templates or support scripts.

Which task matters most?

Choose the task that would hurt most if delayed. This might be a client reply, a support triage entry, a content draft, an invoice follow-up, or an internal summary. Avoid trying to protect every AI workflow at once. Pick the one that appears most often and carries the most pressure. If you can keep that task moving, the rest of the day stays calmer.

How do you keep the prompt usable?

A versioned prompt is a prompt with a name, a purpose, and a date. It should include the task, the constraints, the tone, and the output format. If you use the same prompt every week, save it as a reusable block. If you change it, keep the old version nearby. The point is that you can copy the right prompt without thinking while the outage is happening.

Which model can take the rough work?

The fallback model is the next available tool that can do a rough version of the job. It only needs to be different enough that it works when your primary option is down. Test it before the outage, so you know where the button is and what kind of output to expect.

What do you do by hand?

The manual checklist is the part that keeps you from freezing. It should be short enough to use under pressure. Three to five steps is usually enough. For a client reply, it might be: acknowledge the issue, state what you know, state what you need, and set a next step. For a support entry, it might be: identify the account, summarize the problem, choose the safe response, and flag the follow-up. The checklist is the floor that keeps you professional while the tool is down.

How long do you allow?

The time budget is the most important field. If the fallback cannot produce something usable in five minutes, stop and do the manual checklist. This rule protects your hours from the most expensive kind of waiting: the kind where you keep checking, re-prompting, and hoping. A short, imperfect answer sent on time is usually better than a perfect answer sent after the client has moved on.

Keep the fallback alive

A fallback that is never used becomes a fantasy. Review the card periodically, or after any major change in your work. Update the prompt if your task changes. Update the fallback model if your tools change. Update the manual checklist if the minimum useful work changes. The card should feel slightly boring, because it is a small piece of infrastructure that keeps your one-person company running when the AI does not. The point is not perfection.

When the outage ends, keep the card. It is the quiet mechanism that protects the hour. The next time the lights flicker, you will be following a short, calm path that costs minutes, with hours protected.