I can't write code. I built it anyway.

For most of my life, that sentence would have been a confession.

I want to stay on that sentence a moment. Not knowing how to code meant you stood on the outside of the thing you wanted to build and waited for someone who did. Maybe you could describe your idea well enough that they'd build it for you. Maybe you couldn't afford them. Either way, the wall was real, and I lived on the wrong side of it for years. That wall is gone now. I'm proof it's gone — but not for the reason most people will assume.

Here's the assumption. People hear "she built software without coding" and picture me typing a wish into a chatbot and receiving a finished product in return, gift-wrapped. That is not what happened. If you try it that way, you will get a confident mess back, and you will feel worse about yourself than when you started. I know, because producing that confident mess was most of the work. The finished thing came out of everything I did after it.

A boss who can't type.

What actually happened is that I ran the project like a boss who can't type. I couldn't write the code, but I could tell when the work was wrong. So I made the AI prove every claim it handed me. When it gave me a number, I asked how it knew. When it wanted me to approve its own homework, I made it show the receipts. More than once, it told me — warmly, confidently, in complete sentences — that everything was finished and everything was fine. And it wasn't. And I caught it. Not because I'm technical. Because I refused to accept the words "it's done" as proof that it was done.

That refusal is the entire skill. Not coding. Knowing what "good" looks like, and making the machine prove it got there before you believe it.

One job: guesses can't act.

So let me tell you what the thing I built actually does, in plain words, because the plainness is the point. I call it the Memory Firewall. It does one job. It stops an AI from treating its own guesses as facts. When an AI writes something down to remember later, the firewall makes it tag what that claim really is: something it verified, something it was merely told, or something it simply guessed. And here is the rule at the center of it — a guess is not allowed to drive an action. It can be stored. It can be discussed. But it cannot quietly become the reason something happens, until something real confirms it. A guess written on Monday does not get to wake up on Wednesday wearing the costume of a fact.

Hallucination has a human name.

There's a word for the habit I was building the machine out of. The popular one is hallucination. That's when an AI states something false with total confidence — not because it's lying, but because it genuinely cannot feel the difference between what it checked and what it filled in.

The word I keep returning to is older, and it was never a computer word at all. Confabulation. Doctors used it long before any of this. It describes a person whose mind closes a gap with a confident, coherent story they completely believe. There's no intent to deceive. They are not lying. They simply cannot feel the seam between what happened and what they assembled, after the fact, to make the world make sense.

And that is where this stops being a story about machines.

We confabulate constantly.

We are fed information all day and we act on it as fact without ever checking it. Someone doesn't answer a message, and within seconds we've written the whole story of what it means about us. We're certain a plan will work — not because we tested it, but because we need it to. We repeat things "everyone knows" that we have never once verified. And underneath all of it runs the uncomfortable engine: a great deal of what we believe, we believe because it feels good, or because it protects us, not because it is true. We fill our own gaps with confident stories. Same as the machine. We just don't have anything standing between the guess and the action.

Now hold those two things next to each other, because this is the part I can't stop thinking about. We are furious at AI for doing the exact thing we do all day long. We want the machine to check before it speaks. We want it to admit what it doesn't know. We want it to change its answer the moment the facts change. And we would very much like to be personally excused from all three.

There's a blunt word for demanding a standard from someone else that you won't hold yourself to. It's hypocrisy, and I'm not aiming it at you — I'm including myself, because I've done every version of it. And there's a deeper word for why the machine's habit gets under our skin in the first place. It's projection. The confident guessing irritates us in the AI precisely because it is ours. We built the machine in our own image, and we handed it our worst habit right alongside our best ones, and now we flinch at the reflection.

That's what the firewall really is, once you strip the software away. It's a mirror we built for the machine and then had to look into ourselves. The tool forces the AI to do the one thing almost none of us do without help. It has to separate what it knows from what it's only assuming, call the guess a guess out loud, and refuse to act until something real backs it up. Reading that sentence, most people feel a small sting of recognition. Good. That sting is the whole point.

I didn't set out to fix machines.

I'll be honest about the order I learned this in, because it matters. I did not set out to fix machines. I spent years trying to build this exact discipline for myself. I wanted to stop acting on comfortable stories, and to make my own beliefs earn their keep before I bet my time or money on them. The AI version came second. It turned out the machine and I had the very same problem, dressed in different clothes. I built the machine's fix first for one plain reason: a machine will hold still long enough to let you install one. A person won't. The install in your own head is the harder job, and I am still in the middle of mine.

The same fight, aimed at writing.

The firewall isn't the only thing I built out of this one idea, though. It's just the one aimed at what an AI treats as true. There's a second instrument, aimed at how an AI writes — and unlike the firewall, you can use it this minute, with no technical anything at all.

I call it the Dual Lane Writing Method. Here's the problem it solves. AI writing has a signature failure: it comes out fluent. Smooth, confident, grammatically flawless, and completely hollow — paragraphs that carry the shape of meaning with none of the weight. You've read a thousand sentences like that. They have a shape where a point should be. And if that failure sounds familiar, it should. It's the same failure the firewall fights, aimed at a different target: a guess dressed as a fact, fluency dressed as substance — a confident surface with nothing underneath.

Dual Lane is the check for the writing version. It scores a draft on two lanes at once. Lane One asks the machine's question: is this clear, is it ordered, does it actually say something? Lane Two asks the human one: does this sound like a person wrote it for a person, or like a brochure wrote it for nobody? The whole method fits on one line — the structure must hold, and the human texture must breathe. A draft isn't finished until it does both. You paste your writing in, and it tells you which lane is failing and shows you the exact phrases that gave it away. Nothing to install. Nothing to learn.

I want to be clear about one thing, because it's the first place people's minds go: Dual Lane is not a trick for sneaking AI writing past a detector. It's the opposite. It's an instrument for making writing genuinely better — clearer, more grounded, more honestly yours — instead of merely smooth. And here's the proof, the kind I keep insisting on: the essay you are reading right now was run through it before you saw it. If a sentence in here reads like a brochure, I missed one. The standard is public, so you can check my work — which is the whole spirit of both these tools.

Two kinds of builders.

If you're like me — you have something you care about getting right, and you've spent your life assuming the building part belonged to someone else — it doesn't anymore. You can start today, and you don't even have to start with code. Paste your next important email into Dual Lane. Put the firewall's habit on your own thinking. The only thing to learn is the one habit both tools are made of: don't trust the output, make it prove itself.

And if you're one of the people who can code, who's been doing this for years and does it well — you can build on top of both of them. I made them to be foundations, not finish lines. The woman who built them couldn't write either one by hand, and that isn't a disclaimer. It's the point. If the standard can come from someone who lived outside the wall, then the wall was never the thing that mattered.

Only what I can prove.

Let me end by telling you what the firewall isn't, because refusing to oversell it is the only honest way to describe a tool like this. It isn't finished. Thousands of people are not using it yet. I am not going to call something true because it would sound good in a post. What I will tell you is only what I can prove. It exists, it's tested, the code and receipts are public, and the person who built it couldn't write a line of it herself.

If that's possible — really sit with what you've been telling yourself you can't do. Notice how confident that story was. And notice how little you ever checked it.