Skip to main content
← Back to Source Journal

The Work Comes Back as a Mirror of Whoever Made It

22 July 2026 · Ben Visser · 5 min read

At a glance - TL;DR

Ben reflects on a stretch of unusually clear building time, opened up once a pile of heavy decisions (the largest being whether to take outside money) finally settled. Back to sourcing the company's AI agents himself, hands in the work, he arrives at a realisation he could only reach from inside it: an AI hands your whole way of working back to you, so the same task given to two people comes back as two different pieces of work. The year and a half before this was not falling short but the apprenticeship that made the understanding possible, and he holds it as the latest thing the work has handed him rather than a destination reached.

Key takeaways

  • Working with an AI is closer to briefing an employee than operating a tool: the whole skill is in communicating clearly what the work has to be.
  • Hand that briefing through another person and you stack two rounds of instruction, and the truth of what you hold gets thinner at each step.
  • An AI hands your whole way of working back to you, so the same task given to two people comes back as two different pieces of work.
  • Getting good work out of it is less about the tricks of prompting and more about knowing what you are, because whatever you are is what comes back.
  • A slow stretch can be the apprenticeship an understanding needed; some things only arrive after enough hours with your hands in the work.

Hey there,

Wednesday morning is the one I keep. The rest of the week fills the way weeks do. But Wednesday I block out and sit down properly, the week's notes in front of me, to look at what actually happened. Small realisations turn up all week, always in the middle of something, and in the middle of something there is never room to stop with them. So they wait. Wednesday is where I finally stop for them. I want to be plain about how that stopping happens now, because it matters to the whole of what I want to say. I do it in conversation now, thinking out loud with an AI, my notes beside me. The looking gets sharper as we talk. A year ago that sentence would have embarrassed me to write. It doesn't now, and understanding why is more or less the entry.

What the Clearing Gave Back

For months I had a small pile of heavy decisions sitting open at the back of my mind. The largest was whether to take outside money into the company. Over these past weeks they settled, one by one, until the pile was gone. I did not expect how much room that would leave. With the questions answered, I have had whole days with nothing pulling at my attention, and I have spent them building.

That is the part worth being honest about. We build AI agents, software that does the slow work of cybersecurity compliance. This last stretch I have been building ours myself, directly, hands in it. Not scoping it and handing it down. Sitting with the work and making it, one long stretch after another, the kind of day where you look up and the afternoon has gone. I have not worked like that in a long time. It took the clearing to see how much I had missed it, and how much faster the thing goes when I am inside it rather than beside it.

Two Sets of Hands

Building this way taught me the plain thing first. You treat the AI as an employee, not a tool. Not something you operate and tidy up after, but something that will do the work, all of it, if you can say clearly enough what the work is. The whole skill is in the communicating. And that is why an older instinct quietly broke. I used to hand the task to someone who would then instruct the AI for me. I am the one who holds what our compliance work demands, and what an auditor would need in order to trust it. Pass that through another person on its way to being built and you stack two rounds of instruction on top of each other. What I hold gets a little thinner at each step, a little truth lost in translation twice over. No failing of anyone's, just one retelling more than the work could bear. At first, doing it myself was only me stopping that leak.

But doing it myself set up the thing I actually want to write down. I could only see it once the work was in my own hands. I put the agents I had built next to ones a colleague had built for the same product. We had aimed at more or less the same thing. They had come out completely different. Not one right and one wrong. I want to be careful to say that, because a comparison is easy to mistake for a verdict, and this is not one. Mine had come out as guides. They want to ask you where to begin, they lead with a warmth you can hear, they worry about meaning and about not taking a number for the truth behind it. His had come out closer to machines, deciding the route before anything else, exact and precise, careful with every record they touch. Both are good. Both are needed. Each is strongest in exactly the place the other is lightest.

Seeing them side by side showed me something I had missed entirely. The AI does not only take your instructions. It takes your whole way of working and hands it back to you, larger. Everything I make carries my instinct to guide, to lead with warmth, to ask before deciding. Everything he makes carries his. The same task, given to two people, comes back as two different pieces of work. The work is partly a mirror of the person who made it. Once I had seen it I could not unsee it. Getting something good out of this is not about learning the tricks of how to ask. It is about knowing what you are, because whatever you are is what comes back to you. And that is not a thing anyone can tell you from the outside. It shows up only once you have had your hands in the work long enough to watch yourself appear in it.

The Time It Took to Learn

I want to be careful how I tell the year and a half before this. The easy version is that we were falling short and have finally got it right. That is neither true nor fair. We needed that time. Every month of it. There was a whole craft to learn that no one could hand us ready-made. How to work with these tools, how to keep the code in good order. Where the old way of stitching systems together still helped, and where it only got in the way. How to build on a technology that kept changing under our feet as it grew more capable. The early scaffolding we built taught us the ground even as we outgrew it. Enough has come together now that we can drop a great deal of it and go straight at the thing. But we could only drop it because we had built it first, and learned what it was standing in for.

The deeper lesson took just as long, and could not have come sooner. That you treat the AI as an employee. That what you are is what it hands back. None of it is something anyone can talk you into believing. I had read the words. I could have said them all at the beginning and not understood a thing I was saying. It arrives only after enough hours with your hands in the work. Enough that came back wrong because you were vague, enough that came back right because you were clear, until it becomes something you know, not something you were told.

So there is no regret in any of it. I keep checking myself on that, because I distrust the story that calls every slow stretch secretly essential after the fact. But this one holds up. Take those months away and I would be standing here now with all the room to build and none of the understanding to build well.

What I notice, sitting here on a Wednesday with the afternoon ahead of me, is not that I have arrived anywhere. It is that the work keeps handing me things I could never have been told, and this one, that whatever I am is what comes back to me, is only the latest. A year ago it would have been something else. A year from now it will be something else again, something I cannot reach for yet, because I have not put in the hours that would let me hold it. Where that goes I don't know. And the plainest pleasure of it is this. Even sitting down to understand my own week, I now do the same as everything else that is working: clearly, out loud, with something that meets me and hands back whatever I bring.

With care, Ben