What I bring to the machine

In my last essay I argued that AI output is only as good as its user. An argument like that is easy to aim at other people, so it seems only fair to turn it on myself. If the output is only as good as the user, then everything the machine produces for me is only as good as I am, and I should be able to say honestly what that is worth. That does not come naturally to me. I have never been comfortable talking about what I have done, and at work I tend to let the results speak rather than speak for them myself, so treat what follows as a reluctant case for the defence.

I have always described AI as a kind of superpower built on the skills you already have, and the flip side of that belief is uncomfortable. Without real ability underneath, you are not getting superpowers. You are just announcing nonsense, only faster and more fluently. Someone can take thin work from an AI, show it to people outside their field, be told it is brilliant, and believe it, right up until they walk into a room of people who actually know the subject and the floor gives way. The uncomfortable truth is that in some rooms, I am that person. I am not an engineer, yet I have built software that a church runs its Sundays on. To the people who use it, it simply works. I have no illusion that an experienced engineer would look at it the same way, and I would rather say that myself than have someone else discover it for me.

So the honest answer to what I bring depends on which room I am standing in. In engineering and design I am an outsider, and the machine is doing far more of the work than I am. In product, operations, money and people, I have spent years, and there the relationship runs the other way. Most of what I believe about AI comes from noticing how differently it behaves for me in those two kinds of rooms.

In the rooms where I am an outsider, my first rule is not to trust my own applause. I know the work looks right to me partly because I can't see what is wrong with it, so I borrow depth from wherever I can find it. I get second opinions, from another model, from tests and from people who actually know the craft, and I keep asking questions until I understand enough to decide, rather than shipping what I can't explain. I treat those rooms as a classroom. My ambition there is not to become an engineer, but to learn enough to ask better questions and to notice when something is off.

The other rooms are where the case for my skills really sits. I have been doing product work since 2020, and by now I can usually tell within a paragraph whether a plan will survive contact with the people who have to build it, which means I can tell the same thing about a plan an AI hands me. Years of running operations across very different programmes have made me almost stubborn about where the truth lives, which is why, when I lost access to my Claude account earlier this year along with much of the context I had built up inside it, my answer was not to find a more trustworthy provider. It was to make sure no provider ever holds what I depend on again, so my context now lives in plain files I own. And working inside regulated finance has left me with a reflex that no tool has ever had to teach me: before anything that commits me or someone else's money goes anywhere, I look at it myself.

The machine can tell me what is possible. Knowing what is appropriate has always been my job.

People are the room I think about most, and it is the one where the machine feels most like an outsider to me. I help run a mentorship programme for young men, and an AI can draft a message to a whole cohort in seconds. What it cannot know is which line will land wrong with someone who is going through a difficult month, which joke works with this group and not the last one, or when something should not be sent at all. That knowledge comes from years of showing up for the same people, and it is exactly the kind of depth the machine cannot borrow from me unless I bring it.

If I had to name the thread running through all of it, it would be a mindset more than a skill. When the floor gives way, there are really only two ways to respond: let it humble you and go deeper, or go looking for easier rooms where the applause is safer. I know which one I want to be, and most days I think I manage it. Choosing the humbling over the applause is also what keeps me holding my own standard rather than the machine's. Left to itself, AI will happily hand you the average and call it finished, and the easiest room of all is the one where you accept that average because nobody around you can tell the difference.

So here is my honest verdict on the user. In a few areas I have gone deep enough to reach what my mentor calls the edge, the place where the machine's knowledge thins out and I can see what it is missing. In others I am still in the crowd, borrowing other people's depth and trying not to mistake fluency for skill. The machine has let me work far beyond my own limitations, and my aim is to get something like a hundred times more out of the same hours. But if my own argument is right, the only way to get there is to keep raising what I bring: going deeper in the fields I know, and learning enough in the ones I don't to tell good work from bad. The machine multiplies whatever I bring. My job is to keep bringing more.