A note on humans, machines, and the work only one of them can do that made me reflect after reading this thread – https://x.com/hilbertspaess/status/2097476196791709843?s=20
Humans made AI. It was not the other way around. I say this not as a slogan but as a reminder to myself, because on some mornings, staring at a dashboard that has already sorted, tagged and summarised a thousand messages before I have finished my tea, it is easy to forget who built the dashboard, and why.
I have spent twenty-two years explaining government to people. Helplines, notices, press briefings, the long nights before a big announcement when a single wrong word can reach a few crore homes by breakfast. For most of those years the work was done by hand, by tired people, in rooms with too little light. AI has changed that. It has changed it more in the last three years than in the previous nineteen. And I am grateful for it, in the way you are grateful for a colleague who quietly takes the worst part of the job off your desk.
But here is the thing I keep coming back to. AI does not run a process from end to end on its own. It cannot. Somebody has to decide that it will. Somebody has to sit down, look at every step between a citizen’s question and the answer they receive, and say: this one the machine can do, and this one, and this one too. The machine does not automate itself. We automate it, one step at a time, and at some point, if we are not paying attention, there is no step left with a human in it.
That is the alarm. Not the robot with the red eyes. The slow, reasonable, budget-friendly removal of every place where a person used to be.
I want to be precise about what I am not saying. I am not saying slow down. I am not nostalgic for the days when a helpline operator typed the same reassurance four hundred times in a shift and went home with nothing left for her own family. That was not dignity. That was fatigue dressed up as service. If a machine can take the four hundred repetitions, let it. Let it draft the first version of the notice. Let it read the overnight mail and tell me which twelve messages actually need me. Let it translate, format, schedule, sort. This is what the technology is for: to do away with the daily routine that wears people down, so that people can do the part that only people can do.
And what is that part? It is the moment when the four hundred and first caller is not asking a routine question. She is a mother whose son has just walked out of an examination hall and she does not yet know whether he is all right. There is no template for that. There is a voice, and a pause, and the willingness to stay on the line a little longer than the script allows. The machine can find her file. It cannot sit with her. Empathy where it counts is not a soft skill. In public service it is the whole point, and it does not scale the way software does. It scales only through people who have enough left in them at the end of the day to give it.
So when we deploy AI, the question I ask is not ‘how much can this replace?’ It is ‘what does this free my people to do?’ If the honest answer is ‘nothing, we are just cutting the headcount’, then we have not modernised anything. We have taken a tool built to relieve human beings and pointed it at them. AI should not be deployed to eliminate humans. It should be deployed to return humans to the work they were hired for and never had time to do.
There is a harder truth underneath this, and it is about us, not about the machine. The onus is on the people in the room, the ones with the authority to sign the order, to be smart enough to know when it is time to step aside. Step aside from the tasks we no longer need to guard, the formatting and the sorting and the first drafts, and stop pretending our value lives there. Our value never lived there. It lived in judgement, in the phone call nobody wanted to make, in knowing which sentence would calm a crowd and which would inflame it. Stepping aside from the routine is not surrender. It is the only way to step forward into the part that matters.
But knowing when to step aside also means knowing when not to. It means keeping one human hand on the process at the points where a wrong answer is not a typo but a life. It means refusing the last mile of automation when the last mile is a person in distress. The same intelligence that tells us to hand over the drudgery must tell us where to stop, and that intelligence is ours. Nobody is going to build it into the software for us.
We made these tools. They carry our intent, our shortcuts, our blind spots, and our better instincts too. What they become in a public institution will not be decided by the vendor or the model. It will be decided by whether the people who run the institution remember what the work is actually for.
I remember it best at the end of a long day, when the machine has done its four hundred repetitions and I have done my one conversation that mattered. That is the division of labour I want. That is the one worth fighting for.