Is Your AI Model Going Offline a Risk?
Your preferred language model going offline is not a new risk. But now that you are automating things and using agents, your approach to this risk may need rethinking.
An AI service going offline sounds scary, but it is a familiar risk. SMEs have lived with it since they moved email, files and apps to the cloud. Most of us could cope with a short interruption without the internet, or Microsoft 365. Automating work with AI carries the same risk, perhaps with one difference.
When a cloud service goes down, work stops cleanly and your files are still there when the service returns. An AI automation, by contrast, can stop halfway through a task or process, and you are left wondering where it stopped. The issue is compounded when your team realises that everyone has been relying on this automation for so long that no-one can remember what the process was, or how to manually pick it back up.
The good news is that your automation no longer has to be built around one model. If you are set up properly, it reduces this single point of failure. Your instructions, your process steps and the knowledge the AI draws on (its ‘memory’) can live in your own files and systems in plain formats that you can read and control. As crazy as this sounds given all the attention they are getting, the model is an easily replaceable part. If your current model goes down or changes, another model can pick up the same work from the same place. Perhaps it is time to unbundle your business from OpenAI or Anthropic.
I recommend you give this risk some attention next time you are thinking about risks in your business, or doing some continuity planning. It needs some preparation to keep clear oversight over your processes, and to start thinking about how you would ‘hot swap’ a language model. The goal is to make an AI outage look much like a cloud outage: an inconvenience your business already knows how to manage. Start thinking about this now before the first major outage hits.