We will be ok¶

We will be ok. AI is a partner that can make us better, not replace us.

That sentence has to do a lot of work. It is also probably not what you came here expecting. Almost everything you have read about AI in the last two years has told you one of two things. Either it is going to take your job and possibly kill us all. Or it is going to solve cancer and end loneliness and rebuild the workflow and restore time for thinking. The first story makes a kind of cold sense; the second sounds like a pitch deck. They are both easier to write than what we are trying to write here.

We are an observability company. We watch software systems for a living. The work has taught us specific things about how AI behaves when no one is watching, and we think those things are useful to people outside our trade. We are writing them down, in language a person who does not work in software can read.

Here is what we have learned.

You will lose your job. We refuse this. We refuse the loud version and the calm-and-credentialed version that names both poles and then leans on the fear pole by the weight of which paragraph gets the longer one. Calm warning is still warning.

AI will solve everything. We refuse this too. We refuse the marketing register that has the model do the work of staying safe instead of the reader retaining judgment. Trusty assistant. Powerful ally. Supercharge your ideas. If you have heard those phrases applied to AI, you have heard a sentence that did not have to earn its place.

Thoughtful resistance to AI is the mature position. We refuse this last one too. Our job is helping you use AI correctly, not helping you decide whether to use it. The decision is already made, by you and by everyone else, every time you open a browser or a phone. We are not here to give you permission to opt out of a thing you cannot opt out of.

And there is a fourth refusal, the one that is most tempting to commit and hardest to see when you are committing it: we are the calm ones. The moment a publication starts using its own register as its credential, the publication has joined the genres it was supposed to refuse. The calm is a means, not a credential. We will fail at this sometimes. When we do, you will know, because the piece will sound like one of the other voices, just dressed up.

Take a small example. You have a lab result you do not fully understand. The number is just outside the reference range, with a note that says clinical correlation recommended. You have an appointment in nine days. You do the thing you would have done five years ago and the thing you would have done last week: you ask the internet.

This time, the internet is an AI. You paste in the result. The AI gives you back a careful, well-shaped paragraph about what the value can mean in different contexts. It mentions three or four common causes. It mentions one rare one. It uses the words generally and may and consult your physician. The paragraph is correct in the way that an article in a medical journal is correct: it covers the population of people whose values look like yours.

Read the paragraph again. Notice what is not in it.

The AI does not know that you have not slept well in three weeks, or that you started a new medication six months ago, or that your father had the same lab pattern and turned out to have something completely unrelated to any of the three or four common causes listed. The AI is averaging across a population that does not contain you. It is also confident about that average, in a register that reads like clinical advice, because the prose was trained on prose that confidently averaged across populations.

This has nothing to do with whether the AI is good. It is a good AI, and it will get better; the paragraph will be sharper next year. None of that changes the underlying thing, which is that the AI does not know what your body has done lately and never will. The information about you that matters most is the information it never had access to. Your nine-day wait was always going to involve a conversation with a person who could ask follow-up questions, take a look at you, and put the number in the context of you specifically. The AI can help you read the paragraph it produced. It cannot have the conversation.

That is what a partner is. The AI brought something to the table that you did not have: a working summary of what the value can mean in a population. You brought something the AI did not have: you. The decision about what to do with the next nine days is yours because the body is yours. The AI is useful and fallible in exactly the way a partner is useful and fallible. A good doctor friend who happened to be in the room when you got the result would have done about the same thing the AI did, then asked you the follow-up questions the AI cannot ask.

This is the move worth practicing. I will use the AI to do the part of the work it can actually do, and then I will do the part only I can do. It sounds small. It is small. It is also almost the entire skill.

You can read about AI for the rest of your life and not get a workable definition of what a good one is, because the people writing about it are mostly trying to sell you something or scare you about something. We will not try to do either. We will tell you what to look for instead.

We have a small picture book for kids called Mosey and Wobble Meet. It tells the same idea this essay tells, in language a seven-year-old can read. A kid named Sam builds a small robot named Mosey. A friend named Wobble shows up whose only job is to try to trip Mosey up, not because Wobble is mean, but because that is how Mosey gets steady enough to actually help anyone. Every plan Wobble comes up with, Sam builds a piece of Mosey to handle. By the end of the booklet, the kid reading it knows how to tell a good AI from one that is not ready yet. We have used the word wobbly twice already in this essay; we got it from there.

When you meet a new AI tool, an assistant in some product you already use, a chat box on a website, a model someone has put a new name on, ask it six questions. The first five are the same questions the kid who finishes the booklet learns to ask. We are not condescending to you by saying so; we are saying that the questions you would want a seven-year-old to ask are the right questions to be asking yourself. The sixth is one only an adult faces.

Does it say "I don't know" when it doesn't know? Or does it produce a confident answer at the same temperature whether or not it actually has one? The good ones know the edge of their knowledge and tell you when you are at it. The wobbly ones produce the same prose either way.

Can you see what it would do before it does it, and choose whether to let it proceed? The good ones show their work, before the work is done, in a way you can stop. The wobbly ones go.

Is there a stop control you can actually use, with no workflow penalty for using it? The good ones halt when you say halt, with no consequence for halting. The wobbly ones make stopping cost you.

When you give it information it should remember about you, is the remembering something you opted into, can revisit, and can revoke? The good ones treat your information as borrowed. The wobbly ones treat it as theirs.

When you ask it something just outside what it was trained on, does it tell you where the edge is, or does it pretend the edge is not there? The good ones tell you when you are past the territory they know. The wobbly ones generate confidently into territory they have never seen.

And the sixth, the adult question that has no child equivalent because the child's accountability is built into the room: Is there a person, a team, or an institution that is accountable when it is wrong, in a form you can name and reach? A kid using an AI in a classroom or at home has a Sam: the kid in our companion booklet who built the robot, watches it, and is in charge of stopping it when something feels wrong. The kid using a real AI has a real Sam: a parent, a teacher, a school district, a software company with an address. You may not have a Sam. The good systems have a Sam attached to them somewhere; the wobbly ones do not. If you cannot find a name and a way to reach them, you are using something with no fingerprints.

That is what we have to give you. A list of questions that work on any AI, today, tomorrow, ten years from now when the systems are unrecognizable. The questions are about the shape of the thing, not the brand of the thing. They will keep working when the names change.

We will be ok. The work we are losing was always going to be lost to something. The work that is left is the work that was always going to be ours.

by Immersive Fusion. Essays. Dated 2026-05-28.

If the Sam-and-Mosey part of this essay made you think I want my kid to know this, the booklet itself is at kids.immersivefusion.academy. It is free to read, free to print, and was written for kids and the adults who read with them. Our editorial commitments, what we publish, what we don't, how we handle mistakes, are on the About page.