The way things are going these days, that sentence is currently supposed to sound like a confession. Like I’ve been trying to hide something that’s not polite to mention in public or be proud of.

Except I don’t think I have anything to confess.

Last week, Substack introduced a feature that allows readers to scan an article and receive an estimate of how much of it may have been written or assisted by AI. It has also given creators the option to explain how they work - or disable the scanner altogether. Inevitably, much of the resulting conversation has made that explanation sound more like a confession. The response has been interesting, unsurprisingly polarised and deeply reductive – “Philistine!” “Luddite!” Pick a side and argue to the end.

This week I got a live demonstration of exactly that on my own LinkedIn post - a shorter piece built around the same idea as this one. It was one of my most engaged with posts for some time, with pages of replies: building the argument further, one person testing an AI detector on their own writing just to see what it found, another comparing notes on ADHD and what a genuine thinking partner actually does - and, a few threads over, a flat verdict that the whole thing was "slop," no specifics offered and a beautifully blunt “AI:DR” (although clearly you did otherwise why would you respond?).

There’s a specific sound that goes with the reductive version of that argument. I call it the sigh of ignorance - the small, involuntary exhale of someone who’s just learned AI touched a piece of work, and who has decided, before asking a single question, that it must have been phoned in rather than sweated over. 

Given how much AI-assisted work genuinely has been phoned in, that's not an unreasonable bet. It isn't malicious. It's just lazy - doing exactly what it accuses the work of: skipping the effort of finding out where the craft actually lives and what AI actually brings to it.

Thankfully, Substack also invited writers to explain how they make their work.

And that, dear reader, is far more interesting than arguing about whether AI was or was not used. Let me tell you why.

A percentage might tell you that one algorithm has detected the fingerprints of another. It cannot tell you anything that actually matters about how the piece was made.

Whether I used AI is the least interesting question about this article because, whether you’ve followed me for a few weeks or for the last decade, you already know the answer: of course I did.

The questions that matter are how I used it, what I refused to hand over and whether the finished work is better because of the partnership.

So let me show you the working.

The ritual of authorship

This isn’t the first time we have confused the tools used to create something with the value of the thing being created.

Back when I was at college, studying one of the UK’s first Information Technology degrees, in the late 1980s and early 1990s, I wrote my assignments on a Macintosh SE. For those who weren’t there, this was a small grey box containing what felt at the time like the entire future of human civilisation.

I could move paragraphs around without retyping them. I could correct mistakes without covering the page in Tippex. I could think, edit and refine as I went. It changed everything about how the work got made.

Until of course the work had to be introduced to the outside world, a world where computers were the exception not the norm. The Poly owned room after room of computers: mostly beige Packard Bell boxes running WordPerfect on orange screens, connected to shrieking dot-matrix printers. And still my esteemed educators insisted that assignments be submitted in handwriting. No exceptions.

So I would compose the work on my Mac, print it out and then copy the whole thing by hand onto sheets of paper, swearing with every word and secretly plotting a career spent helping people adopt new tools without losing what matters.

The ideas on the paper were mine. The research was mine. The argument was mine. But apparently the work only became authentically mine once I had made it harder to read by scratching some ink over a bit of paper.

How we all laugh now at the thought of the futility of that approach, and yet, I’ve a funny feeling we might be right back here again.

A new capability arrives. People begin using it because it helps them do something better. Institutions (and people) distrust it because the existing rules were designed for the world before it existed. So the tool gets pushed into the shadows while everyone performs the old ritual in public.

The technology arrives first. The norms take longer.

What is actually mine?

The starting point for everything I write is mine.

The initial observation, the thing I have observed that has been irritating me, intriguing me or refusing to leave me alone, starts with me. So does the point of view I bring to it.

I have spent more than three decades thinking about the relationship between humans and technology. I believe we repeatedly apply twenty-first-century tools to nineteenth-century thinking. I believe effectiveness matters before efficiency. And you must know by now, that above all else I believe the point of technology is to elevate human capability rather than simply remove human cost.

Those aren’t positions generated for whichever article I happen to be writing that week. They are the long arc of my work, evidenced through articles, books, keynotes and the occasional pub rant.

Over the past year, I have put much of that work into a private AI library: words I’ve written or spoken in keynote transcripts, articles, unfinished arguments, research and the stories I have collected along the way.

None of that is especially unusual. Lots of people are building systems capable of copying their style and helping them generate content. Mine can do that too, and it is extraordinarily useful. But that was never the primary goal. It is really just a by-product of what I wanted to build: a partner that understood my arguments and could help make my work better.

It knows the stories I return to and, more importantly, why I return to them. It knows what I have previously said about productivity, creativity, automation, skills and accountability. It can show me when a new idea fits naturally into that body of work, when it contradicts something I already believe, or when two arguments I have kept in separate drawers might actually belong together.

I call it my “second brain”: a place that does a better job than I can alone of remembering not just what I’ve said, but the complex relationships between it all. It frequently makes connections and associations I might never have found, simply because no human brain (and certainly not mine) can hold the full breadth and depth of thirty years of thinking and experience in working memory at once.

The conversation before the writing

Before I start an article, I use AI to explore the idea. I pull at threads. I test different interpretations. I ask what I might be missing. I ask how someone who disagrees with me would respond. I look for the point where an interesting observation becomes an argument worth making.

Together, we might sketch the structure: where the story should begin, which ideas need to arrive in which order and where the conclusion has to land.

Sometimes I write the first draft. Sometimes the machine does. Often, we take turns.

It will suggest sentences I dislike, arguments I reject and metaphors that should never be allowed outside the machine that created them. Occasionally, it offers a line that expresses exactly what I have been trying to say. More often, it gives me something close enough that I can finally see what the sentence should have been.

The authorship does not lie in who typed the first arrangement of words.

It lies in who decided what the piece was trying to say, who judged which words belonged, who removed the things that were inappropriately glib, generic or wrong and who decided that the finished result accurately represented what they believed.

Nothing is published because the machine produced it. It is published because I chose it.

Turning experience into raw material

Every keynote I give is recorded and transcribed.

A transcript that simply sits in a folder is a log, and a log is dead weight. Instead, I use those transcripts as raw material for the next conversation.

What landed in the room that I had underplayed in the script? What did an audience question expose that I hadn’t properly considered? Which offhand line turned out to be the actual idea?

The corpus isn’t just a record of what I have said. It is the raw material for what I haven’t yet worked out.

AI helps me find those moments across hundreds of thousands of words. It can notice that an argument I improvised in Dundee connects with a question someone asked in Barcelona and a paragraph I wrote ten years ago.

I might eventually have found that connection myself. But as a gentleman of a certain age, I also might not.

That is one of the great opportunities of this technology. It does not merely help us create new material. It can help us see more value in the experience, knowledge and unfinished thinking we already possess.

The uncomfortable work

Some of the most valuable conversations are the least comfortable.

I have asked the same system to examine the market for people doing work similar to mine and tell me where I am genuinely distinctive and where I am simply adding another voice to an already crowded conversation.

What am I known for? What should I be known for? Where is my argument strong? Where am I relying on familiar stories because they are comfortable? Which subjects have I neglected because they are difficult?

A human adviser can do this brilliantly, but the conversation is inevitably shaped by a relationship. An algorithm has no need to preserve my feelings or protect its next invitation to dinner. I’ll tell you another time about Claude’s brutally honest assessment of my middle-aged weight gain. For now, just know that it happened.

That does not make it automatically right. It does make it a useful mirror. And, as with every mirror, the responsibility for deciding what to do with the reflection remains mine.

The point was never more

The visible result of all this is that I can now commit to publishing two thoughtful pieces every week - something I could never have sustained over the past decade because of the time and cognitive commitment it would have required.

But cadence is the least interesting benefit. The point was never simply to create more. We already live in a world with far more content than any of us could possibly consume. Producing another thousand competent words is not, by itself, a meaningful contribution to society.

The point is to make the work better.

Sharper arguments. Weaknesses found before the reader finds them. Connections made between ideas that would otherwise remain separated. Research surfaced when it matters. Familiar assumptions challenged before they harden into lazy conclusions.

The ambition is not more work. It is work that lands north of where I would have reached on my own - and that, my friends, is my entire argument for the difference between using AI for efficiency and using it for effectiveness.

What the machine cannot do

AI can present me with options. It cannot decide what I mean. It can identify a contradiction. It cannot decide which side I should take. It can tell me that a paragraph is unclear. It cannot care whether I mislead you. It can mimic conviction. It cannot possess any.

It can draw on more words than I will read in a lifetime, but it has never stood on a stage and felt an audience disengage or laugh at moments where I had not yet seen the joke. It can anticipate how a line might land. It cannot be the one who answers for it when it lands wrong.

Those things matter.

The machine brings scale, memory, speed and an extraordinary capacity for pattern recognition. I bring purpose, experience, judgement and accountability (and occasionally an entirely out-of-band word like spatula, thrown in to weaken the signal, ever so slightly, for whichever future algorithm eventually eats this article).

Neither contribution is insignificant. They are simply different.

Beyond the purity test

I understand why people are protective of wholly human creation. That instinct is defending something real. Some art derives part of its value from knowing that another human made it alone, from inside their own experience.

But I also suspect we are approaching AI’s “Judas” moment: the point when a revered analogue creator publicly plugs in the machine and part of the audience cries betrayal. As with Dylan, we will eventually realise that the art was never contained in the machine being used. It came from the human using it.

Not everything should be AI-assisted. Plenty of our best work never will be, and should not be. But that does not mean all assisted creation is somehow diminished.

A writer who uses an editor is still a writer. A photographer who uses photoshop is still a photographer. A musician using a synthesiser has not delegated the music to electricity.

Tools change the process. They do not automatically determine the value of the outcome.

AI can make lazy people lazier and creative people more creative. It can enable someone to produce large quantities of work without thinking, and it can help someone else think more deeply than they could have managed alone.

The difference is not simply whether the tool was present.

It is the purpose with which it was used, the judgment applied to its output and the accountability retained by the person whose name appears at the top.

An AI detector may be able to estimate whether a machine touched the prose. It cannot distinguish between someone publishing the first answer a chatbot produced and someone spending hours challenging, rejecting, restructuring and rewriting its contribution.

It cannot tell you who had the idea, who cared whether it was true or who will answer for it when it is wrong. That is not a criticism of the detector but instead is the reason transparency matters.

So yes, I write with AI. Gladly, deliberately and transparently.

The arguments are mine. The judgment is mine. The decision to publish is mine. And when I get something wrong, the responsibility is mine too.

And that, surely, is a much more useful measure of authorship than whether I moved a pen across a page - or placed every word there unaided.