AI / Wanderings 2026
AI Will Expose Our Values Before It Improves Our Work
Andrew Forrest’s Fortescue interview is a useful reminder that artificial intelligence makes company values more important, not less.
I have just spent 43 minutes listening to Andrew Forrest, the Chairman and CEO of Fortescue, talk to Nicolai Tangen.
There were the expected things: iron ore, a railway, a cyclone, a broken leg, green energy, a billion litres of diesel that Fortescue wants to remove from its supply chain.
There was also artificial intelligence. Smart grids. Autonomous trucks. The prospect of a renewable-energy system reacting so quickly to a fault that the lights do not even flicker.
Impressive stuff.
Yet the sentence that stayed with me had nothing to do with technology.
“Never rely on contracts. Rely on people.”
Forrest said it while reflecting on Anaconda Nickel, an earlier business that hit serious problems despite the comfort of a big engineering contract. Contracts mattered. The people mattered more.
That is a useful thought for the age of AI.
We are currently treating artificial intelligence as if it were mainly a software procurement exercise. Buy a licence. Write a policy. Ask everyone to complete a 20-minute training module. Put a cheerful chatbot on the intranet. Congratulations, you have transformed.
No, you have acquired software.
Transformation is what happens when that software meets the culture you already have. And AI is brutally good at exposing culture.
It will expose whether people are allowed to say “I do not know”. It will expose whether a manager values judgement or simply wants answers to arrive faster. It will expose whether safety is a real operating principle or a poster next to the fire extinguisher.
AI does not create those questions. It makes them harder to avoid.
Humility is suddenly an operating system
Fortescue has ten values. Forrest was asked which one mattered most. He refused to choose a favourite child, but he singled out humility as especially protective.
That is exactly right.
AI can be wonderfully useful. It drafts first versions, finds patterns in a mountain of information, spots faults on a grid and removes some of the administrative sludge from a working day. I use it constantly.
It can also sound utterly convinced while being wrong.
That is not a small bug. It is the central leadership challenge.
The organisations that do well with AI will create a culture in which someone can say: “This answer looks plausible, but what is it based on?” Nobody gets punished for asking or labelled difficult because they slowed down the shiny machine for thirty seconds.
Humility gives us that pause.
Humility reminds us that a confident answer is not the same thing as understanding. A junior employee may challenge the output; a senior leader still carries moral responsibility for the decision.
Dear executive committee: if the only person allowed to question an AI-generated recommendation is the person who bought the licence, you have not built an AI strategy. You have built a very expensive echo chamber.
Hit the crocodile closest to your canoe
One of Fortescue’s values is courage and determination. Forrest explains it with a very Australian instruction: “Hit the crocodile closest to your canoe.”
Solve the immediate problem. Stay on course. Do not get distracted by the dramatic thing on the horizon while the thing beside you is already taking on water.
This should be written above every AI steering committee.
Most companies do not have an AI problem. They have a process problem, a data problem, or a leadership problem. Occasionally the full collection, which is a charming way to spend the budget.
The closest crocodile is usually not whether an AI agent will replace half the company by Christmas. It is more ordinary:
- A customer-service team spends hours every week hunting for information that should be easy to find.
- A clinician is drowning in documentation and has less time for the person in front of them.
- A school is teaching pupils to produce answers while a machine can already produce very polished answers in two seconds.
Start there.
Pick a painful, human problem. Make the work better. Measure whether it actually helped. Then carry on.
Forrest calls this a “crazy brave Plan A” with a “bulletproof Plan B”. That is a far better recipe than launching 37 pilots and hoping one of them becomes a strategy.
Generating ideas is easy. Keeping good people is harder.
There is another Fortescue value that has become strangely relevant: generating ideas.
AI has made idea generation cheap. You can ask for 50 campaign concepts before your coffee goes cold. You can generate a business plan, a product name, a spreadsheet model, an apology to a customer, and probably a haiku about the quarterly forecast. The haiku will be decent. This is not the point.
The scarce things are now judgement, taste, responsibility, and the courage to say that an average idea is still average when it has been formatted beautifully.
Forrest makes a harder point when discussing people who perform brilliantly but damage everyone around them. Keep them, he says, and they make sure nobody else performs.
AI will amplify this problem.
With an AI assistant, a controlling manager can create more control at speed. A vain executive can publish more certainty. A healthy team with permission to experiment can learn far more quickly.
Same technology. Very different company.
This is why values cannot live in the annual report alone. They need to show up in the workflow, in the review of an output, in ownership of a decision, and in the consequences when it goes wrong.
Safety now includes the human mind
Fortescue puts safety first. In mining, that makes immediate sense. Huge vehicles, remote sites, physical risk. Nobody needs a keynote speech to understand why a safety value matters near a 400-tonne truck.
AI has its own version of the 400-tonne truck.
It can quietly alter how people think, decide, learn and trust. An employee who stops checking the work. A child who stops struggling with a difficult question. A public service that cannot explain why a person was denied support. A company that optimises a metric and forgets the human being attached to it.
Safety in AI means protecting data, of course. It also means protecting human agency.
We should design systems that leave room for people to understand, intervene and disagree. The friction is not always a defect. Sometimes it is the part that keeps us awake.
Be useful. Enjoy it.
At the end of the interview, Forrest recalls advice from an old Australian stockman: “Live your life usefully. Now, enjoy it too, young fella.”
That is better than most AI manifestos I have read.
AI should make us more useful. A nurse ought to gain more time with patients. A small business should be able to understand its numbers without paying an oracle in a suit. A teacher deserves more time for curiosity and less for policing formatting. An engineer should be able to fix the grid before the lights go out.
And yes, it should be enjoyable. Useful work has a different energy. You can feel it in a team that is building something real, learning quickly, and treating people properly while doing it.
The future will not be decided by whoever has the cleverest model in a demo. Models will get cheaper and features will be copied. Slides will continue to be slides.
The durable advantage will be the human one: a group of people who know what they stand for, who can use powerful tools without becoming servants to them, and who remember that technology is here to help us do better work for other people.
That is the crocodile worth hitting first.
This post was inspired by Nicolai Tangen’s interview with Andrew Forrest on In Good Company.