I was only nine when the storm known in insurance circles as 87-J put the lights out in millions of homes across France and England, and knocked down six of the big old trees at Seven Oaks. I learned something that day about the wisdom of remembering how we did things in the ‘olden days’.

My mother pulled some plain candles from the back of a drawer, and my father, armed with a torch, retrieved an oil lamp and a paraffin stove from the garden shed.

“When I was a boy, this was our everyday heat and light,” papa said. “It’s important to remember how we used to get things done when things go wrong.” Even then I was aware he was casting himself back much further in time than corresponded to his actual experience (I had been into his childhood home, which even in his day had then-modern electricity and a ‘town gas’ cooker). But the lesson was not wasted. It is an important one to heed today, at home and in the office.

I probably shouldn’t be writing this in a technology column, but no matter how many precautions are built in, tech will inevitably fail, if only for a brief spell. This week’s CloudFlare outage provides yet another headline example. Downtime stopped thousands, maybe millions of businesses from trading for a few hours. A longer interruption would have caused serious red-ink leakage from P&Ls world-wide.

When IT outages happen, we need to be able to do things the old fashioned way. Remember CrowdStrike? The outage took out a widely used airline staff-scheduling system. But one carrier got their boats back into the air the faster than all the others, because their HR manager used to do the schedule with a pen and paper, remembered how, and had showed the team. As soon as the cloud-native solution went down, they reverted to manual. Passengers flocked to the only airline in the skies, so for them, the strike was only a blip.

Systems are revolutionising the insurance market, but if we rely on them entirely, without a Plan B, we may lose all the productivity gains they deliver. I don’t want to come across as a paranoid survivalist, or worse, a flat-Earther, but we are becoming dependent on fallible systems. We need to:

• know how to calculate a rate if the underwriting workbench overheats
• remember how to trade in person if the algorithmic exchange goes down
• be able to make a table of risks and claims
• understand that the supercomputer in our pocket can be used to make phone calls, allowing us to actually speak to our business partners, as well as to send them virtual notes.

On a practical level, we need to remember to teach people about the processes behind these essential functions. But my fear about our increasingly utter reliance on technology goes profoundly beyond a concern that we will forget how to act when the tech breaks down. Please indulge my worry.

In the modern, AI-energised workplace, thought is too often replaced by the output of AI tools which, as I reported last week, are self-admittedly only “optimized to approximate truthful, helpful answers,” rather than hardwired to deliver the actual truth (and they go down, too: ChatGPT crashed with CloudFlare).

My sons used to argue that using Wikipedia for homework support was no different than using the old encyclopaedia chez papa. This is wrong for two reasons. First, their homework was always in English, and the Encyclopædia Universalis was in French. More subtly, looking things up in the old books required more than simply typing a word in a box, and cut-and-paste was an impossibility. They had to think and to write, actions which underpin learning.

Nowadays, in the GenAI age, copying from Wikipedia looks far preferable as a learning exercise than asking ChatGPT to approximate a convincing middle-school essay.

My company and many others make insurance technology that is changing the nature of our business. However every company should ensure that they engender an institutional memory of tech-disabled processes. If we don’t, we will forget how to cook dinner when the power goes out, and someone else will eat our lunch.