AI is speeding everything up. Software development, data analysis, the everyday work of running a business - all of it moves faster than it did a couple of years ago. That is mostly good news. But speed has a habit of creating problems somewhere else. In India we joke that a new flyover does not remove the traffic jam; it moves it to the other end of the flyover.

So it is worth asking what else has to change when everything gets faster. My answer, perhaps an unexpected one, is kindness.

Think about the roads. When we travelled by horse and cart, the streets were dirty but fairly safe. Nothing moved very fast, so people could walk on the road without much risk. Then the motor car arrived, and with it, speed. Speed meant that drivers had to be careful around the people still walking. It also meant new laws, new road designs, new safety standards and driving tests. None of these were needed for horses.

There were two ways to respond. The first was to keep the car and take away the speed. Britain actually tried this. The Locomotive Act of 1865 limited powered vehicles to a few miles an hour and required someone to walk in front waving a red flag. You get a new engine, and then you are told to go no faster than the horse. It did not last.

The second way was to let people drive fast, but teach them how. Train them, license them, and expect them to mind their road manners around everyone else. That is the path we took, and it is the path we need with AI.

Speed is safe only when it comes with consideration

Here is where it already shows. We increasingly let AI summarise what people have done, write their status reports, assess their performance, even coach them. People are flawed, and their work rarely tells the whole story. A colleague who was there knows that the project slipped because a supplier went quiet, or that someone was covering for a sick teammate. AI does not know any of this. It is not unkind by nature; it simply lacks the context to empathise. And without that context, its judgement can be much harsher than a person’s would be.

Recruitment is the clearest example. When AI started screening CVs, we heard story after story of the most suitable candidates being rejected. The model was not being cruel. It just had no way of seeing what a human recruiter would have read between the lines.

There is another response, also borrowed from the roads. We built motorways, where walking is not allowed and everyone has to be in a vehicle. The AI equivalent is that if I have to deal with your AI, I bring my own. That works a bit like mobile phones. One person with a mobile phone is not much use; it only becomes seamless when both people have one. So mobile phones took over, and then we had to learn to live with them. Ringtones at weddings. Calls in the middle of meetings. Over time we built an etiquette - no phones at the dinner table, no calls on the Tube - because the technology made it easy to be inconsiderate, and we had to learn to be kind on purpose.

AI will need its own etiquette. But there is one big difference. Cars and phones were tools; a person was always in charge of them. AI agents act on their own. And more and more, we will not know who is on the other side. A message arrives with a name and an email address, and we assume a person wrote it. It may well have been an agent.

So the question is not only how we behave around AI. It is how the agent behaves around us. An agent cannot be considerate unless we teach it to be. This goes beyond giving a model a friendly personality or a pleasant tone. It is about reasoning with kindness - noticing that there is a person on the other end, allowing for what it does not know, and giving them the benefit of the doubt. Whether that belongs in the training of the model, or in how we design the agents around it, is an open question. It quickly drifts into philosophy and morality, and I do not think we have the answer yet.

My guess is that we will end up with different kinds of models for different jobs. Some will be strictly factual and objective, and that is exactly what we want for reconciling accounts or analysing data. But agents that deal with people, and with human situations, need something more. They need to be taught how to be considerate.

We learnt to drive fast, and then we learnt to drive with care. Our agents will need to learn both.