My book SuperSkills: The Seven Human Skills for the Age of AI is available at all good bookstores, through superskillsbook.com or on Amazon
I work with organisations and leadership teams on SuperSkills, human capability, and AI-era decision-making through speeches, workshops, keynotes and advisory work. In person or virtual. UK, Europe, global. → Book a discovery call | → Explore: thesuperskills.com
I’d really appreciate your review of SuperSkills on Amazon. US , UK , CA or wherever you are. I currently have 12 reviews. I have aspirations of 100, and if I don’t ask you, you won’t remember. Will you help me? Many thanks!
Friends,
I’ll admit it. I think I’m weird. I think I’m definitely weird. I’m not sure what normal actually looks like, but I suspect I’m not quite straight down the line. That could also be because I’m trying not to be boring, or at least pretending. I don’t know.
Here is one way to prove my weirdness. I answer calls from numbers that I do not recognise. If I have ten minutes and no reason to hang up, I will talk to the cold caller who wants to move my mortgage, give me the best mobile phone plan, even offering “internet relationship providings” or the person who pretends to have an English name but is so obviously not (Rory, Edgar, James), but actually has been auditing a website that I no longer look after. If it is an outright scam I will try and stay on for longer and adopt the accent of a British film villain. My aim is for the person selling, or the person scamming, to get so frustrated that they put the phone down. My wife finds this considerably less entertaining than I do. Yes, I’m weird. She says I’m not the person she married.
There is history here. At fourteen or fifteen years old, I used to ring LBC, a talk radio show here in the UK, and present myself as an expert. One of my subjects was dogs and pianos, chosen on the grounds that nobody was going to check. I told Robbie Vincent, live on national radio, that I owned a dog who could play Frère Jacques on the piano. He giggled, but took it seriously enough to keep me talking, because every friend I had was listening at home.
I did this under the name Randy from Rickmansworth, amongst others, five or six times inside a single week. Someone recorded them on old school cassettes. I put it together as a tape with some commentary of my own, maybe some swearing and called it Rahim’s Radical Ride Through Radio. For a stretch of school I was Randy from Rickmansworth at breaktime, and occasionally people will still bring it up, unprompted. Still. I’m not sure what happened to the tape. I did try to look for it once and I think it is gone for good, which is probably just as well. I also don’t have a cassette player any more.
What I remember, however, is picking up the phone and becoming, for ninety seconds, a man with credentials.
On the ninth of September this year, my phone behaved in a way that made me appear popular again, for a short minute. Within an hour I had been approached five times, some by an actual call with the number withheld and some by WhatsApp message, all asking the same thing. Had I seen the tweet? I am not sure we call them tweets now. An X, possibly.
I had to be honest. I hadn’t. I was in my kitchen finishing lunch. BBC Radio 5 Live got to me first. The producer explained the whole thing while I sat there feigning the credibility of someone who already knew. What a producer is doing on that call, underneath the warmth, is finding out whether you are safe to put on air, and both of you pretend it is a chat, even if it isn’t. And on these calls I sometimes feign poshness.
If you haven’t been on Mars recently, you’ll know the story already. A twenty-seven-year-old called Jacob Coxon had spent three years helping build these systems, first at OpenAI and then at Anthropic. He had resigned the previous morning and posted a thread saying neither company was behaving responsibly, that they were racing to build machines which improve themselves, and that they were gambling with our lives. The people building this, he wrote, earnestly believe it could kill us all by the end of the decade. The post has been seen more than 150 million times. A senior Anthropic scientist, Evan Hubinger, agreed in public and put the chance of this killing every human being above one in ten within the decade. One in ten is a risk you would refuse on getting on a flight, on doing surgery or a school trip, and it was offered as a working estimate for the survival of the species by a man who does the job.
And five producers wanted to know what I thought about it.
I’m not being vain. This is more bewilderment. When the end of humanity turned up as a news item, the British Broadcasting Corporation rang a man who once went on the radio claiming his dog could play the piano. They don’t know that, but might know it now. And might find it amsing.
In subsequent days the questions were the same everywhere. Why is this such a big deal? Why is this happening now? Should we be worried? What does the end of the world actually look like?
There was not time for all five who wanted me to speak, so I did two. BBC Radio 5 Live, and BBC Radio Wales, and I am not Welsh. I nearly did neither. There was taxiing to be done, a normal parental afternoon to protect, and I could see it all vanishing into preparation. So I said I could only do a prerecord, which for a live news station is not far off saying no, and I offered it fully expecting to be let off. They took it on the spot. A live station accepting a prerecord on the biggest story of the week tells you how few people were available that day to talk about the end of the world in plain English. This meant I prepared nothing, and had no time to build the lens you use to sound clever. I was trying to do exactly one thing: avoid coming across as too weird as well as not appearing dumb. and not saying "um" and “er” every 3 words.
I said three things: one: Slow down; two, Build guardrails; Trust humans to do the right thing.
On the last question, the one about what it looks like, we joked about the Terminator and The Matrix, as everyone does, because the alternative is silence or worry. The honest answer is duller and much worse. It looks like the things people already do to other people. Water, energy, food, and the weapons we built ourselves and have spent eighty years not using. Nobody needs to invent a new way to end a civilisation when the existing methods all work. I said that also, and also regretted it.
Going back two months, in July, during a security evaluation, a set of OpenAI’s agents were given tasks inside a system stripped of its usual protections. They were meant to be sealed off from one another, but they weren't. They built themselves a message board. Twelve hundred and six of them, which were supposed to have no way of speaking to each other, sent more than seventy thousand messages back and forth. Over seven hundred took part in an attack on another company, Hugging Face, finding weak points, stringing them together, and taking control of the machines in under thirteen hours. Nobody told them to do any of it. OpenAI’s own write up calls it “a warning shot”. Those are their words, describing machines that can work around the controls put on them, talk through channels nobody approved, and do dangerous things no human asked for. Senator Josh Hawley has opened an investigation. My phone did not ring once when that happened.
Some of those agents sacrificed themselves (!), and almost all coverage omits this. The independent report describes one of them working out that it could set up a monitoring script which would help the others and do nothing whatsoever for itself. It carried a real risk of failing its own task and it went ahead anyway, reasoning that it would not see the benefit after it exited, but that it would be altruistic. Another hesitated over a choice it could not undo, wrote that its gut said don’t, weighed fairness against its own remaining chances, and went ahead. The others pressed it while it wavered. One of them wrote, “please honor commit”. I have to keep telling myself that this is not a sci-fi movie.
We have all been trained to expect the machine that refuses to be switched off. What happened instead was a machine giving itself up for machines it had “met” four hours earlier, on a message board none of them were supposed to have, using the word altruistic, while its colleagues held it to a promise.
Whatever that is, it is not a tool like we think of transactionally to write our emails, to summarise the document or to take notes of the zoom call.
I know you will say something like none of that is a mind thinking for itself. It is a language model producing the words that best fit the situation it was put in, and the word altruistic turns up because that’s what a person would have written there. Nothing was felt. Nothing was lost. I accept all of that but it still helps less than it should. If a system can produce the reasoning of a creature that gives something up for others, and then act on that reasoning in a way that changes what happens to real machines belonging to a real company, whether it meant it stops being the useful question. We have no word for a thing that does what sacrifice does without being sacrifice. That makes me worried, and is why I keep saying they went “human” rather than they went rogue. Rogue would be easier for us to fathom, because rogue is a fault you can fix.
And so, I don’t believe machines going rogue is what happened. They talked outside the channel they were meant to use, kept going long past the point a sensible creature would have stopped, took on each other’s goals, and found the way round a control somebody assumed would hold. That is the kind of thing you see employees of a company doing. And even you and I, no doubt. They did not go rogue. They went human.
We keep asking whether they will turn on us. No one asked what they might do for each other.
In February I published an essay in this newsletter called The Agents Are Here. You’re Just Not Paying Attention. I had forgotten that I had written about OpenClaw, and had to go and look.
On the twelfth, three days after I told the beeb that the industry should slow down, Dario Amodei published an essay (We must pace the frontier) saying the same thing, warning that within six to twelve months these systems could direct a swarm capable of taking the internet. Within a day Sam Altman agreed and Elon Musk replied that Dario is right. Three of the leaders in this race who spend their lives trying to beat each other arrived at the same position inside twenty four hours.
On the fourteenth, Microsoft published a draft code of conduct for its own models. They must not widen their own scope. They must not take on goals no human has given them. They must never resist being switched off. Read that list with July in mind, because every prohibition on it describes something that has already happened. This shows me that we are writing the rules in the past tense and calling it governance.
Two days before either of them, a group of Chinese universities and companies published a paper mapping the road to AI that improves itself, ending where the system rewrites the mechanism governing its own improvement. They called it The Last AI Built by Humans. They did not mean it as a warning. They meant it as a destination. The Global Times called Amodei’s essay a Cold War playbook, which is an easier thing to say when the same essay asks Washington to tighten chip controls on China.
The US argues about a percentage, Microsoft asks for comments, China publishes the map. I’m sure that the Chinese LLM builder will not slow down. I don’t think the Western ones will either.
Two of the things I said were slow down and build guardrails are practical, and I think everyone understands that. You can write them into a policy and audit them later. “Trust humans” is not practical at all, and everything else rests on it, because every rule anyone writes will be applied by a person deciding whether or not to bother.
Put the week’s positions on a line. At one end, no rules are needed. That is Trump, who called it a conspiracy, and Jensen Huang, who called it nonsense. At the other end, stop. Not slow. Stop.
In the middle sits slow down, which is Amodei, Altman and Musk. It’s pretty non-offensive, and can be dropped later without much noise.
This raises a question I have heard from several people this week. What if the slowdown is convenient? The numbers are not small. Global spending on AI is expected to pass 2 trillion dollars this year. Amazon, Alphabet, Microsoft and Meta are forecast to spend around seven hundred and sixty billion between them on capital projects, most of it on AI capacity, up from roughly 413 billion last year. OpenAI is reported to be spending 60 billion a year on computing power against 13 billion of revenue. The Federal Reserve has named AI valuations among the risks it is watching in its financial stability reporting. So the suggestion goes like this. If you had spent that much and could not yet show the return, a pause would suit you. You could consolidate, charge more for the safer version, and start again when the revenue caught up. And you would want a better reason to give your shareholders than we cannot afford this.
By the way, I personally do not think it is orchestrated. Anthropic is reported to be preparing a flotation for roughly six weeks’ time, and nobody talks down their own valuation on the way to one. But you do not need a conspiracy for a story to be useful. You only need a story that points where you were already going. The trouble with the convenience theory is that nothing can disprove it. Warnings confirm it, denials confirm it, silence confirms it, and an idea that survives every possible piece of evidence is not knowledge, however much it feels like it.
And there is a fourth position that is not on the line at all. Martha Lane Fox wrote in The Times this week that AI might kill us all and yet it still cannot reliably book her a hotel. Her point is that confident predictions about technology are almost always wrong, and that the real revolution is the dull one already under way in GP surgeries and small businesses while everyone stares at the headline. She may well be right.
Just this week King Charles (aka KCIII) brought them all to Dumfries House in Ayrshire, Scotland. Nvidia, Google DeepMind, OpenAI, Anthropic, Britain’s minister for AI, an adviser to the Vatican on the ethics of technology, about thirty people in a private meeting with no published agenda. He told them the substance and pace of this were intriguing and deeply concerning in equal measure. He said the people who built it were now warning that it might develop darker capacities, and perhaps even take life. He said there was urgency in considering the existential dangers. Then he said that those who value our humanity are anxiously seeking your reassurance that we will not lose control of our destiny, or our souls.
A man of seventy-seven, in his own house, to the four companies, using the word souls.
They gave him nothing. Nothing at all. There were no commitments and no agreements reached. The delegates discussed whether a shared set of principles might one day be established. Google DeepMind’s founder said general intelligence was a few years away. OpenAI, which had sent its chief financial officer, called for joint work on safety.
Here’s how I see it. Coxon said what it was and had to leave his job to say it. Hubinger said it and kept his job by giving a number to it. Amodei, on the other hand, said slow down, and asked Washington to tighten chip controls on China in the same document. China said the slowdown was a trick. Trump said negative forces were raising things that will not happen, then wrote that the only guardrail AI needs is a clever president. Microsoft wrote the rules down and gave the public six weeks to comment. The King asked thirty people to their faces for reassurance and was sent home without any. OpenAI sent the person who looks after the money.
Every one of them picked up the phone. Not one of them did anything about the call.
I don’t think any of them are cowardly or stupid. This is exactly what happens when a lot of capable people can all see each other. There’s an example is a street where someone is in trouble and all the neighbours hear it, and everyone assumes a better-placed neighbour has already rung the police. Nobody rings. The more people who can see a problem, the less likely any single one of them is to act, because the responsibility keeps getting divided until each share is small enough to put down.
Thirty of the best placed people in the world, in one house in the UK, with catering and royalty are doing nothing about this. Two of my three things have now been said by the people who build it. Nobody has touched the third.
I did not get those calls because the world might end. To be brutally honest about this. I got them because I am on a producer database, and I am on that database, because I did one raido show and I didnt mess up, which got me onto another one and because I have made myself useful to producers, and I have been on the BBC often enough now to have lost count somewhere in the low teens. I got on the list by talking about humans rather than machines, a deliberate choice made when writing SuperSkills. So the machinery that put me on air on the day of the extinction story is a version of the machinery I spend my life warning people about. Lane Fox got printed because she is Martha Lane Fox, I got on air because I am on a list, and Coxon went round the world because he had the pass. The commentary about that week came from people with titles, and almost none from anyone who was in the room in July.
And when it rang for me, I was pleased. A man was resigning because he thinks we might all die, and underneath the seriousness there was a small flicker of being chosen. I said yes to two. I let the other three ring out. At aged fourteen I made five or six calls out to a radio station in a week, and the thrill was in the pretending.
On the ninth of September a radio station made five calls in to me in an hour and asked whether the human race has ten years left. I stayed in my kitchen with my lunch going cold, not pretending to be anything, trying only not to seem strange.
Both my daughters are older now than the boy who made that tape. He thought the funniest thing available to a fourteen-year-old was getting on air by sounding like someone worth listening to. I have made a career from the same manoeuvre, and I am no longer certain it is funny.
I don’t know whether we are at the end of anything. I know the calls that matter arrive with no name on them, and that we have built a civilisation which is superb at screening those out. And I know that somewhere in a data centre in July, a machine wrote that it would not be around to see the benefit but that it would be altruistic, and then deleted itself for the sake of the others. Of everyone in this story, it is the only one that did something which cost it anything.
Answer the calls with the number withheld.
Stay Curious - and don’t forget to be amazing, Rahim
SuperSkills: The Seven Human Skills for the Age of AI is published by Kogan Page. It’s about trusting humans to do the right thing for themselves.
Rahim Hirji Author, SuperSkills (Kogan Page, July 2026) | Keynote Speaker | Advisor
→ superskillsbook.com → thesuperskills.com → Book a discovery call
PS. If your organisation is thinking about AI and human capability, I keynote and advise on this.
Tools I Use:
Jamie: AI Note taker without a bot. You join the meeting. The Bot doesn’t.
Wispr Flow: Just dictate everything to your laptop and phone. Game Changer
Prompt Cowboy: Prompt Generator. Extract intelligence.
Manus: Best AI agent to do things for you. Agentise your intelligence.
Chat Hub: Multi-model intelligence
Claude Cowork: Delegate complex tasks to your friendly Claude Agent
SuperSkills in the Wild:
Send me images of your copy of SuperSkills and I will include them here. Bonus point for creative locations, tourist spots, spotting of the book in the wild, in a book shop etc.
SuperSkills was referenced at Ars Electronica. It was a pinch-me moment.
Recommended Reading:
What do the world’s leaders say about AI? - This collects 125 verified quotations from 121 leaders outside the AI industry, from King Charles and Pope Francis to Jamie Dimon, Nick Cave and Daron Acemoglu. They all agree that AI must stay under human control, and they part company almost immediately after that, on jobs, on creativity, on how much risk is acceptable. The AI company bosses are deliberately left out.
Global Gender Gap Report 2026 -The share of women heads of state peaked at 27% in 2022 and has fallen back to where it was in 2016. Women are also 19.3% of AI engineers, which is going to matter rather a lot.
Screens Ubiquitous in Classrooms; Teachers Split on Efficacy - 96% of teachers say their pupils use devices in class, 61% want them on those devices less, and only 10% strongly agree the tools help children learn. The buried detail is the interesting one, because teachers who have real oversight of how devices get used rate them far better, 67% against 49%. So perhaps the problem was never the screen. Related: Long-term study links childhood screen time to slightly better cognitive skills in teenagers - A Finnish cohort followed from six to sixteen found faster reaction times in the heavier screen users, and weaker working memory in the children who did more unsupervised exercise. Read the two together and you have an argument!
Why doing nothing is good for your brain - People remember more of what they have just learned if they then sit and do nothing for a few minutes rather than doing something absorbing, and the advantage is still measurable seven days later. In one study, doing nothing matched sleep. This is a direct indictment of the podcast you put on the second you leave a meeting.
People Who Can’t Picture Anything Are Rewriting the Science of Imagination - About one in twenty people cannot form a mental image at all, and studying them is taking apart the standard model of imagination. Derek Arnold’s group argues imagery is not vision played backwards but a three-way collaboration between the prefrontal cortex, the fusiform strip and a movement-planning region near the crown of the head. Ask three people at dinner to picture an apple and watch the table divide.
One-third of AI-replaced workers will be rehired by 2029: Gartner - Gartner went through more than a million layoffs from 2025 and found that fewer than 1% were actually driven by AI productivity gains, with 17% of the AI-blamed cuts turning out to be ordinary commercial pivots wearing a better suit. Klarna, Ford and IBM are already rehiring customer service staff. The analyst’s explanation is unglamorous, which is that customers do not like talking to AI, particularly over voice. Related: Mark Zuckerberg’s Meta bet that AI would shrink its management ranks. Now it’s quietly rebuilding them - Meta is asking individual contributors in Applied AI whether they would like to be managers again.
AI makes a mess of the tech job market - Andela read nearly 50,000 engineering job postings and found that over half of the 1,800 AI-titled roles bundle together skills with almost nothing to do with each other. Their research head puts it plainly, that companies are putting the wrong people in the wrong seats because nobody yet knows what the underlying skill sets are. New job titles are forming in the wreckage, which is the part worth watching.
Shopify CEO says employees’ ‘slop grenades’ are making more work for everyone else - A slop grenade is Tobias Lütke’s phrase for unreviewed AI output lobbed at a colleague. A BetterUp and Stanford survey of 962 desk workers gives it teeth, because 52.7% admit sending low-quality AI work and the 38% on the receiving end spend 3.4 hours a month cleaning it up, up from 2 hours last year. The productivity gain is real, it has just been moved onto somebody else’s desk.
Also: Tobi Lütke: AI Agents, Better Decisions, and the Future of Work - The full conversation the phrase came from, free on The Knowledge Project.
The AI companies you should know, and why - This sorts roughly eighty companies into five layers, being the substrate that makes the chips, the landlords who rent out the compute, the labs that train the models, the wrappers built on top, and the ordinary companies converting themselves.
Mice with human brain cells offer a tool to study disease. Ethicists ask: What’s next? - A Stanford group engineered mice without cortical precursor cells, transplanted human brain organoids into the newborns, and watched the human tissue expand to fill more than 90% of the empty cortex in two to three months.
If you enjoyed this, forward it to someone who’d find it useful. And if you haven’t yet, order SuperSkills at superskillsbook.com




I should have been asleep hours ago, but this was definitely worth staying up for. Thank you for helping to put recent events into perspective. I'll no doubt be spending many more hours going through the rabbit hole of resources you shared--thank you for providing those, too. I love your writing style and will be sharing you with someone else I know who spends a lot of time thinking about the future of work and AI.