For thirty years we took the human parts out of work, school and service because screens were cheaper. AI will either finish that job, or hand those parts back to us. Which one happens is a design decision — and almost nobody is making it on purpose.
That decision is our whole business.
Measure one side only and the answer is always the same. We measure both.
The technology is not the variable. Same models, same money, same week — and two completely different countries at the end of it. The fork is chosen at design time, by whoever writes the objective. Flip the switch and read the same three problems twice.
AI is pointed at productivity, so productivity is what improves. The line goes up, the country still feels worse off, and nobody can say why.
AI quietly absorbs the functions a neighbourhood used to perform, until the reasons to leave the house run out one by one.
AI companions substitute for people. Easier every time, and every time a little more permanent.
AI is pointed at wellbeing, and the number that governs spending finally matches the thing voters actually feel.
AI spots who has been missed, then gets a person there. The machine finds it; a human does it. Never the other way round.
AI prepares people for human contact — rehearses the hard conversation, lowers the stakes — then gets out of the way.
None of these are harder to build. They are only harder to ask for — and nobody in the room is paid to ask.
None of them slow anything down. Each one simply refuses to let a technology decision be signed off with only half the ledger filled in.
If you cannot name what people get back from a deployment — in hours, in skill, in contact, in agency — the design isn't finished. Write it next to the ROI, on the same page, at the same board meeting.
Compute lands in a real place, on real land, drawing real water and power. Build the shed, fund the green space, the school lab, the community space of equivalent value. Infrastructure with a visible human dividend attached.
The fastest way to make humans look redundant is to score them like machines. Score the contribution and what they get back from the effort, and the same workforce becomes the reason customers stay.
An hour saved is only a gain if somebody receives it — as learning, development, care, rest or time with the people they love. An hour saved and reabsorbed as more throughput is not efficiency. It's extraction with a dashboard.
Nobody should ever have to guess whether the voice, the tutor, the carer or the colleague is human. Disclosure isn't a compliance box — it's what keeps trust in the humans. The EU AI Act now requires it; we think it should be a point of pride, not a footnote.
Automation of the routine is fine; automation of the exit is not. When the stakes are real — money, health, a grade, a livelihood — a person must be reachable, empowered and able to overrule the system. The off-ramp is part of the product, not an escalation cost.
Pilots still hand-fly. Surgeons still cut. When AI takes over a capability, the humans who oversee it must keep practising it — or the day the system fails, nobody in the building can do the job. Delegation without atrophy is a design requirement.
Any technology that sits between people is scored on the human contact it creates — conversations started, rooms filled, relationships kept — not the hours it captures. Time on screen is a cost. Time with each other is the return.
“Is AI going to hollow out our degrees?” Only if assessment stays where remote learning left it. Move the proof back to interaction — discussion, defence, teaching, judgement — and AI becomes the thing that restored the seminar rather than the thing that killed the essay.
“How much can we automate?” is the wrong first question. The one that decides whether you still have a company in five years: what do we want our people doing with the capacity this creates — and how would anyone outside know we meant it?
Compute is now national infrastructure, approved at planning speed with no human-return test attached. Pair every gigawatt with a named, funded, measurable public gain and the politics of AI stops being a fight about whether it gets built.
Net Human measures the human return on technology, advises the people building and governing it, and certifies the organisations that can prove the number. If you want both sides of the ledger filled in, start with a conversation.