“It’s Just Moving Data” (and Other Things That Aren’t)

The Objection

I have been doing versions of the same work for most of my career, and for most of my career I have been answering the same question. Early on it was data. Why is the migration taking so long, it is just moving data, pick it up here and put it down there. Later it was systems. Is this not overcomplicated, do we really need all of that. Now it is AI. Why does the thing need to check its own work, why does it generate its own guiding documents, why does it need alignment on how it behaves, are you not overbuilding this.

The object keeps changing. The objection does not. And after enough years you stop hearing it as a question about the specific task and start hearing it as one continuous confusion, wearing whatever costume the decade hands it.

The confusion is not stupidity. Some of the people asking are the sharpest people in the room. It is something more interesting than that, and David Hume described it nearly three hundred years ago without ever seeing a database.

The Invisible Cause

Hume noticed something unsettling about how we understand cause and effect. We never actually perceive necessity itself. We see one thing, then another, and through repetition the mind learns to expect the second from the first. The billiard ball strikes the second ball, the second ball moves, and we say the first caused the second. But we did not observe the causing. We observed a before and an after, and the connection between them is something the mind supplies, not something the eye was given.

Hold that next to a piece of work done well.

The client sees the data sitting in the old system. Then the client sees it living correctly in the new one. What the client does not see, because it was never available to be seen, is the reconciliation of three contradictory definitions of a single field, the discovery that two departments had been quietly meaning different things by the word "closed" for six years, the decision about which of two conflicting records was the truth, and the two hundred smaller decisions like it. Get one of those decisions wrong and a client's entire matter history can end up silently attached to the wrong file. The client will never see that the decision was there to be made. They will only ever see that it went right, or one day that it went wrong.

So the client does what the mind always does. Sees the before, sees the after, and infers a simple connection, because a simple connection is the only thing the senses were offered. The work was invisible, so the work looks trivial. Not because the client is careless, but because the connection between before and after is precisely the part perception does not hand us. Someone has to reconstruct it.

A lawyer already knows this in their own domain, they have just never had it named. The brief that reads as though the conclusion were obvious is the brief that took the most work. The argument that flows so cleanly the judge nods along is concealing the forty hours of research that got cut, the three lines of reasoning that were tried and abandoned, the single case that reframed the whole thing found at eleven at night. Opposing counsel sees a tidy six pages and thinks it was easy. The ease is the achievement. The work is precisely the part that does not appear on the page.

That is the first move. The work connecting before to after is invisible, so the effort reads as small.

The Habit

Hume's second observation is worse, because it explains why this problem grows instead of shrinking.

We come to expect effects to follow their causes, he said, through nothing more than repetition. See it happen enough times and the expectation hardens into something that feels like a law of nature. The sun has risen every morning of your life, so you do not experience tomorrow's sunrise as a probability. You experience it as a certainty you never think about. The repetition did that. The habit became invisible to itself.

Now apply that to a system that works reliably.

The first time the automated process delivers the right answer, someone is mildly impressed. The tenth time, it is expected. By the hundredth time, the correct answer simply appears, the way the sun comes up, and no one in the building can any longer remember that appearing was ever in question. The reliability you worked hardest to build is the exact thing that trains everyone around it to stop seeing that anything is being done at all.

The better it runs, the more invisible it becomes. The more invisible it becomes, the more it looks like it runs itself. An insurer understands this shape immediately, because they live inside it. The policy that pays out cleanly, quickly, without dispute, looks in the moment like a simple transaction. Money in, claim out. What made it clean was the risk modeled correctly years earlier, the exclusions written with enough precision to prevent an unnecessary fight, the underwriting that priced the thing right before anyone knew it would matter. When it all works, the payout feels routine. Routine is the product. And routine is indistinguishable, to the casual eye, from easy.

So the objection sharpens with time rather than fading. Early on, people doubt the new thing because it is unfamiliar. Later, they doubt the value of the thing precisely because it has become so familiar it disappeared. Both doubts point at the same target from opposite directions.

The Part That Should Worry You

Here is where the diagnosis stops being interesting and becomes your problem. If the reasoning that matters most is hidden from view, and reliability only deepens that invisibility, then good work and bad work can present you with the same surface. At least for a while, both can be quiet. Both can produce an answer that appears on time. The difference is buried in the part no one watches, the reconciliation done carefully or carelessly, the edge case caught or missed, the confident output checked against reality or simply trusted.

You find out which one you bought when it fails, and invisible work fails invisibly too, right up until it fails expensively. A migration with one field quietly mapped wrong across a hundred thousand records. A model that has been confidently wrong for a month in a way no one thought to check. The bill for bad invisible work does not arrive when the work is done. It arrives later, with interest.

The machinery is not the deliverable. The result is. A client should not have to observe every reconciliation, discarded approach, validation check, and judgment call that produced it, and frankly should not want to. Done correctly, all of that vanishes beneath a surface so smooth it feels like nothing happened. That smoothness is the product. It should feel like magic.

But there is a problem hiding inside that success. Good work can leave evidence behind: tests, reconciliations, records of decisions, checks against source material. It should leave whatever evidence the work allows. Evidence, however, is not the same thing as seeing the work happen, and it does not always close the gap. Sometimes, especially with AI, there is no perfect reconstruction of why the system produced exactly what it produced.

So the tension remains. The more completely the difficulty disappears into the result, the better the experience, and the less of that difficulty remains visible to you. The better I do the job, the less the experience itself tells you the job was hard. At some point you are evaluating the result, the evidence that survives, and the person responsible for both. There is no clean way to eliminate that trade. Anyone who claims there is has probably made the demonstration part of the product.

The Close

The question that has followed me across three technologies is really one question. Why does something that comes out simple so often require a process that is anything but simple.

Because the surface cannot tell you how much reasoning it took to make it simple. Simplicity is not evidence that the process was simple. Sometimes it is what careful work leaves behind.

It is just moving data, in the way that a brief is just six pages, or a clean payout is just cutting a check. The "just" is doing an enormous amount of work, and all of that work lives in the connection between before and after, the part perception never gives us directly.

The benefit of an expert is not that you get to watch every decision being made. It is that you do not have to. You get the result, the thing is right, and you are spared the invisible argument that made it so.

That is what expertise looks like when it is working. The best version of complicated work often looks like nothing happened at all.

The work becomes clearest when you imagine what would have happened if nobody had done it.