In August 1896, Bridget Driscoll became the first pedestrian killed by a car in Britain.
At the time, she had been visiting the Crystal Palace, a large exhibition hall that was one of London’s biggest public attractions.
She was crossing Dolphin Terrace with her daughter when she was struck. The car that struck her was giving demonstration rides, driving at only about four miles an hour.
Witnesses said she froze, unsure which way to turn, before she was hit. The coroner later said he hoped “such a thing would never happen again.”
But hope isn’t a strategy.
It would take until 1931, 35 years after Driscoll’s death, for the Highway Code to be published. And the first driving test wouldn’t become compulsory until 1935, 39 years after her death.
We handed people the keys to powerful vehicles before we taught them how to use them safely. That decision came at an enormous cost.
In 1933, there were around 1.8 million cars in the UK, and more than 7,000 people were killed by those cars that year.
In 2024, there were roughly 34 million cars on UK roads, and 1,602 people were killed.
While every death is tragic, that means today’s roads are about eighty times safer than they were in 1933.
I see many leaders making a similar mistake with AI.
They’re handing their teams the keys to powerful tools without making sure they understand the rules of the road.
Many leaders are approaching AI like this:
- They give their team access to a tool on Monday.
- Maybe a lunch-and-learn on Tuesday.
- Then it’s handed off, and the team is left to mostly figure out the rest on their own.
To the leader, it might feel like progress. But it’s creating an enormous risk.
Now, I want to be careful here, because no one dies when your AI makes a mistake. The stakes aren’t the same, and I won’t pretend they are. But the shape is.
A powerful tool in the hands of people who don’t have enough preparation can put your team, or your business, at serious risk.
The good news is that we don’t need to wait for the wreckage to write the rules.
That’s why training on prompts isn’t enough. Prompts are the acceleration pedal.
What people actually need to develop before the prompts is their judgement: when to trust AI, when to push back, what information is safe to give AI, and which parts of the process still belong to them to verify.
That’s harder to teach than the process of using a tool, but it’s the only part that makes the tool safe to use.
If you’re unsure whether your team knows how to use AI safely, here are two questions to ask yourself:
- If someone on your team got a wrong answer from an AI tool, would they notice? And would they feel empowered enough to point it out?
- What have your people actually been trained on, beyond how to log in and start using it?
The moral of the story: we wrote the rules of the road after many, many tragedies.
This time, we know better.
Want more notes from the field?
Melissa shares practical observations on AI, leadership, mindset, and what helps organizations make AI useful.
Give me the Field Notes
