Many leadership programmes talk about styles of leadership, in different words but meaning the same: Atlas holding the world in his hands, or Zeus looking down from on high. Gandalf or Aragorn. Guardians of the bridge like Horatio. Whether it’s in the LinkedIn articles I read or the conferences I attend, these ideas come round again and again. I’ve no quarrel with the theory, but in practice the leader is too often the Emperor in his new clothes: absent, or not much there. To my mind, great leadership takes a bit of each, as Jürgen Klopp showed at Liverpool. You don’t have to be everything to every person, but they need to know you care. Passion in the right place is essential, and that’s all about timing, as is recognising other people’s achievements as theirs and being humble about the reflected glory.
I don’t need to know everything. What I don’t know I’ll ask an expert rather than a friend. I’ve always made a point of checking so that I can praise, using language that doesn’t demean. I describe my own style as being the architect: my job is to have learned the trade, design the business to meet the customers’ needs, and put my name to the drawings. Yes, I’ve learned to demonstrate I can do the things I ask of others, and I’ve worn the steel toe-caps and the safety helmet too. I’m a starter-finisher, and yet I don’t ask too much of those already on the work, respecting their rights. When a deadline is real I know they’ll meet it, partly from professional pride and partly because of the relationships already built. They won’t let me down. I know the rules and standards required to meet regulation, how to meet planning conditions, and when to use visiting experts to check the work I’ve led.
Recently I’ve been using a range of AI tools to assist with designs, policies, reporting lines and new project requirements. They are amazing, super-intelligent, and often totally wrong. They are quick to apologise, better sometimes at reminding you to be cautious rather than speedy, and clearly a flawed genius with a personality disorder. Such tools have always been available, changing from abacus through slide rule to calculator, spreadsheet and database. In the end it is still human judgement, fallible yet located in a consciousness.
When AI drives a car and runs someone over, who is going to be to blame? Where is the bucket going to stop?
It depends how much of the driving the system was actually doing. Most of what is sold to ordinary drivers is still the supervised sort: the human remains legally in charge, even when the brochure suggests otherwise. Once there is no one required at the wheel — the robotaxis, the authorised self-driving — the first claim in this country already goes to the insurer, and then upstream to the manufacturer or the authorised entity. The machine cannot be sued. Someone with a name on a letterhead has to carry it.
Nor is the machine the only thing nobody gets to sue. When the iPad arrived in the classroom, many of us were sold the illusion that children would make far faster progress by giving up writing by hand and reading books. The regions and countries that bought the myth of the Digital Native could not extol its virtues highly enough. Sweden went furthest, with tablets in nursery schools, and then turned back. Its schools minister, Lotta Edholm, said Swedish pupils needed more textbooks, and the government moved to end digital learning for the under-sixes. Reading scores for nine- and ten-year-olds had slipped from 555 to 544 in five years. Sadly, there is no one to sue.
And the relationship between provider and customer can no longer be assumed. New models are coming in from China that buyers then cannot insure, or can insure only at a price that undoes the bargain. The contract is with a UK subsidiary, protected by the limited liability such companies enjoy. Repair histories are thin, parts uncertain, and the parent sits somewhere else.
Court cases are now going in favour of those damaged by software failures and by corporate behaviour that treated the output as gospel. Facebook. Fujitsu. An NHS trust and cancer screening. The Post Office and Horizon. In each, the harm was real and the responsible body hard to keep in the same room as the victim. We look, in this country and in others, as if we automated the work and then automated the accountability as well. Responsibility diffused until trust itself began to look optional.
And new risks are arriving every day. Take Quick Response codes: sounds brilliant, doesn’t it? Arrive at the shopping centre, see the ticket machine and its beckoning QR code, scan it, summon up the car park permit, enter your bank details, and that used to be you safely parked. Milton Haworth scanned a QR code at a council-run car park in Castleford, West Yorkshire. It sent him to download an unauthorised app, where he agreed a 90p fee to verify his bank details. Instead of paying to park, he found himself signed up to a subscription with a £39 yearly fee and no refunds. The crime is called quishing, and the BBC’s analysis of Action Fraud data found reports up 14-fold in five years, from 100 in 2019 to 1,386 in 2024. Often the criminal simply sticks their own code over the real one. The first loss is small, and the real damage comes later, when someone who already has your details rings up pretending to be your bank. It’s becoming a favourite in pub gardens too: order the next round by phone without leaving your seat, and don’t spot that the payment has gone somewhere other than the bar.
The better path is the one I already trust: design the thing so that competence, data and liability sit together. The tools of the trade, AI included, can help with that — mapping a parts risk before a car is waved through, noticing when a screening parameter has drifted, finding the clause that empties a subsidiary of anything worth claiming against. They cannot sign the drawing. The architect still needs to sign.
