The tools got smart. Most of what matters did not change at all.

I build Ai software for a living now, which means I spend my days at the edge of what these systems can do. That edge moves fast. Every few months something that was impossible becomes routine. It is genuinely thrilling, and it is genuinely disorienting, and it has taught me one thing above all others. The hardest part of building with Ai is knowing what to leave alone.

What Actually Changes

Start with the honest list, because the changes are real and pretending otherwise helps no one.

The cost of certain kinds of thinking has collapsed. Drafting, summarizing, translating, searching, pattern-finding across huge piles of text. Work that used to take a skilled person hours now takes seconds. That is not hype. I watch it happen daily. When the cost of a capability falls that far, you do not just do the old thing faster. You do things you would never have attempted, because they were not worth the hours before.

The shape of a lot of jobs changes too. Not “Ai takes the job.” That framing is lazy. It is more that the boring middle of many jobs, the mechanical part, gets absorbed, and what is left is the part that needed a human all along: judgment, taste, the decision about what is worth doing at all.

The question is no longer “can the machine do this.” Increasingly it can. The question is “should it, here, for this person, in this moment.” That question is still ours.

And the speed of building changes. A small team with good tools can now do what used to take a large one. That is real leverage, and it is why I moved into this work. But speed is a multiplier, and a multiplier does not care what it multiplies. Point it at something thoughtless and you get thoughtlessness at scale.

The temptation that comes with the power

Here is the trap, and I have fallen into it. When a tool can do something, you feel a pull to let it, simply because it can. The capability creates its own justification. You automate a customer interaction because you can, not because the customer is better served. You generate the content because it is cheap, not because anyone needed it written.

That pull is the thing to watch. Capability is not permission. The fact that the machine can write the condolence note does not mean it should.

What Does Not Change

Now the part that gets forgotten in every wave of new technology, and this one most of all.

People still want to be understood by other people. A customer with a real problem does not want the fastest possible response. They want to feel that someone grasped their situation and cared about the outcome. Ai can help you understand them faster and serve them better. It cannot want the outcome on your behalf. That caring is not a feature you can ship. It is a posture the humans behind the product either hold or do not.

Trust is still built the slow way. Hannah Arendt wrote about the human condition that action is fundamentally about beginning something new among other people, and that it always carries the weight of being unpredictable and irreversible.1 Ai does not remove that weight. If anything it raises the stakes, because you can now act at a scale and speed that used to be impossible, and the consequences land on real people just as hard. A system that makes a mistake for one person made it quietly. A system that makes it for a million made it before lunch.

And meaning still comes from doing work that matters to someone. Automate the meaning out of a person’s job and you have not helped them, even if you made them more efficient. Drucker’s oldest question still governs: what is the business actually for, and who is it for.2 Ai changes how you serve the answer. It does not change the need to have one.

The line I hold

So I build with a simple rule. Use Ai to remove the toil and amplify the human, never to replace the human where the human was the point.

Use it to draft, so a person can spend their energy on the judgment. Use it to surface the pattern, so a person can make the call. Use it to handle the volume, so a person is free for the moments that need a person. The direction of the tool should always be toward giving people more room to be human, not less.

When I catch myself designing an experience that removes the human where the human mattered, I stop. Not for ethics-in-the-abstract. For a concrete reason: those are the products people quietly come to resent, and resentment is a slow death for anything you build.

How to Build This Way

Three habits keep me honest, and I offer them because they are portable to any team.

Ask who is on the other end. Before you automate anything, name the actual person who will receive it, and ask whether this makes their experience more human or less. If you cannot picture them, you are building for a metric, not a person, and it will show.

Keep a human in the loop where the stakes are real. Low-stakes, high-volume, easy to verify: let the machine run. Correctness-critical, emotionally weighted, hard to undo: a person decides, with the machine assisting. The AI Act and most serious governance frameworks land in roughly the same place, and the instinct behind them is sound. Match the level of human oversight to the level of consequence.

Treat speed as a responsibility, not a prize. You can build faster than ever. That means you can also break things faster than ever, for more people. Slow down exactly where a mistake would land hardest. The discipline is not to move slowly everywhere. It is to know precisely where speed is a gift and where it is a hazard.

I moved from banking to building Ai because I believe these tools can genuinely help people, and I have seen them do it. But the help is not in the tool. It is in the choices of the people wielding it. The machine got smart. Whether the product is humane is still entirely up to you.

So before you ship the next clever thing the model can do, ask the only question that has never changed. Does this serve the person on the other end, or just impress the person building it? You already know when the answer is the second one. Build the first one instead.

Footnotes

  1. Hannah Arendt, The Human Condition (University of Chicago Press, 1958). Her treatment of action as natality, and its inherent unpredictability and irreversibility, sits at the center of the book’s argument. ↩

  2. Peter F. Drucker, The Practice of Management (Harper & Row, 1954), where he argues the purpose of a business is to create a customer. ↩