Sometime last March, while clearing out my Dropbox, I stumbled upon a folder I had nearly forgotten — unpublished blogs and articles I had written between 2008 and 2018. Most had never seen daylight. I had always intended to refine them, publish them someday. Life, training schedules, and the relentless pace of work had other plans.
Reading through them, I was surprised by how much I had actually written — and how much of it was unfinished. But buried inside that sprawl, I found a dozen pieces that shared a thread, a longer narrative arc quietly waiting to be woven together. After nearly two decades of busyness, I finally made the time. I began sieving through them, looking for the spine of something larger.
My thinking had evolved significantly in the intervening years. There was a time when I searched for patterns in history — connecting disparate events, wondering if some invisible hand guided human progress across millennia. But as my understanding deepened, I began to see something else: that order can emerge from chaos, given enough time and enough permutations across the vast expanse of space. Complexity arising not from design, but from the basic laws of physics themselves — iterating, selecting, building. That idea became the gravitational centre of the book. The Turing Threshold was born.
The Unexpected Collaborator
While drafting the book, I found myself doing something I hadn’t anticipated: spending the better part of a year in conversation with LLMs.
Not to have the book written for me. That distinction matters enormously — and I’ll return to it. But to refine my thinking and find the best articulation for ideas I already held.
At first, it felt purely instrumental — a more responsive search engine, a faster way to test phrasing. But something shifted over time. These models did not know anything in any meaningful sense. They did not understand. But they reflected structure. They mirrored reasoning back at me. And in doing so, they exposed gaps — in my logic, in my assumptions, in the coherence of my arguments. I began to think of them less as tools and more as a kind of cognitive surface: a place where messy ideation could be tested, compressed, and refined.
Not intelligence replacing intelligence. Intelligence interacting with its own externalization.
The Low-Effort Problem
Around this same time, I noticed a different pattern playing out in the wider world — a quiet but growing resistance to anything that felt machine-assisted. A disdain for a certain cadence, a certain vocabulary, a certain smoothness that had become recognisable as the signature of AI-generated text.
I understood the reaction. Today’s LLMs, used carelessly, are Garbage-In-Garbage-Out — but amplified. Their output arrives beautifully wrapped, remarkably confident-sounding, and often completely hollow. The polish is real. The thinking behind it, frequently, is not.
This troubled me. Because I wanted to use these tools — but I did not want to produce that kind of work.
First Draft: 372,340 Words
I began cautiously. For the first four months — April through August 2025 — I used only a locally deployed model (Gemma-3-27B) for grammar corrections, one paragraph at a time. The ideas, the structure, the arguments: entirely my own.
The result was a manuscript of 372,340 words. Roughly 1,600 pages.
Reading back through it, the problems were obvious. The narrative swung between philosophy and deep technical exposition, with tangents stretching into what I can only describe as techno-spirituality. The writing was dense — long sentences that unspooled into paragraphs, technical jargon saturating passages that were meant for general readers. I had written a treatise when I wanted to write a book.
English is not my native language, though I have used it long enough that I now think in it more readily than in my mother tongue — with a vocabulary shaped almost entirely by decades of reading in electronics and computer science. That background was showing in every paragraph. What I needed was compression: the ability to express ideas with precision, without distortion, and with the kind of clarity that does not require the reader to have an engineering degree.
A Conscious Choice
So I made a deliberate decision: to externalize articulation while maintaining complete ownership of thinking.
To let the machine assist with expression. To remain fully responsible for meaning.
I began working with a broader array of models — ChatGPT, Gemini, Grok, DeepSeek — feeding in sections of the manuscript, compressing them without losing the narrative arc, refining cadence, maintaining consistency of voice, testing the logic of my arguments, sharpening the ideas under pressure. It became a tight feedback loop.
Readers will spot the tells — the em dashes, the staccato rhythm, the “not this, but that” constructions. I’m aware of them. As a reader of technical and non-fiction writing, I was drawn to this style when it was handled well: sharp, compressed, intellectually alive. The question was whether I could steer it rather than be steered by it.
What the Loop Revealed
Something interesting happened inside that feedback cycle.
The ideas became sharper. Weak assumptions surfaced faster — not because the model identified them directly, but because trying to compress an argument forced me to confront whether it actually held together. Arguments had to survive pressure. Tangents that felt meaningful in isolation became obviously dispensable when I had to explain why they belonged.
It felt less like outsourcing thought and more like placing thought in a feedback system — a part of it externalized, tested, returned. In a strange way, this was the book’s own thesis playing out in real time: memory extended outside the body, reasoning distributed into machines, selection acting on ideas rather than organisms.
The book became an experiment in the very process it was describing.
Not a Mind. Not an Agent. A Surface.
I want to be precise about what I think these tools are and are not — because the conversation around AI is cluttered with overclaiming on both sides.
LLMs are not minds. They are not agents in any meaningful sense. They do not have goals, understanding, or awareness. What they are — when used with genuine human thinking behind them — is something more interesting and more modest: a surface on which cognition can iterate.
We are not building thinking machines. We are building environments in which thought itself can be tested, shaped, and scaled. That is a different claim, and I think it is the more honest one. It is also, I believe, the more consequential one — because it places the responsibility squarely where it belongs: with the human doing the thinking.
The machine reflects. The human must mean it.
An Unexpected Consequence
After nearly a year of working this closely with LLMs, something else happened — something I had not anticipated and find genuinely curious.
My own natural writing began to shift.
The structures the models tend to generate — compressed, rhythmically deliberate, architecturally clean — began to align with how I write when I’m not using them at all. My unassisted drafts started to look, in certain ways, like the output of the feedback loop. The style had internalized.
I would not be surprised if my purely original work were flagged by an AI detector today. That raises questions I don’t have clean answers to — about authorship, about voice, about what it means for a tool to change the person using it rather than just the output. These are not hypothetical concerns. They are happening, quietly, to anyone who works closely with these systems for long enough.
The Dialogue Continues
The book is out now. I hope it resonates.
But this part of the journey — the process, the questions it raised, the strange loop of human intuition and machine articulation — feels like the beginning of something, not a conclusion.
We are in an early and consequential moment. The tools are powerful. The temptation to let them do the thinking is real and understandable. But thinking is not the bottleneck they solve. Expression, compression, iteration — these are where they genuinely help, when the human behind them is actually doing the work.
The distinction between using a machine to think and using a machine to articulate what you think — that distinction is everything. It is the difference between intellectual laziness dressed up in polished language and genuine ideas finding their clearest form.
I made a choice to stay on the right side of that line. Whether I succeeded is for readers to judge.
The Turing Threshold is now available. If any of this resonates — the ideas, the process, or the questions it leaves open — I’d welcome the conversation.