How I turned 45 years of thinking into a 50k-word book in six days
A few days ago, I found myself eating a bowl of cereal while dictating part of a book into my computer and wondering how quickly I was actually talking. I checked with ChatGPT, which put normal conversational speech at around 150 words a minute, so even with the cereal I was probably somewhere close to that. When I am properly in flow I suspect it is faster.
The strange thing is that dictation itself has become a skill. I am learning how to hold a sentence while another thought arrives, when to follow an unexpected connection and when to pull myself back before I disappear down another rabbit hole.
That might sound like a fairly minor observation, but it is actually at the heart of how I managed to write a book in six days. Not because I suddenly learned how to think faster. I didn’t. The book contains more than 45 years of thinking: observations from childhood, lessons from thirty years of building companies and making things, ideas that came from conversations, mistakes, successes, failures and long periods of solitude where I had little choice but to spend a lot of time thinking about people, systems and behaviour.
Some of the ideas have been sitting in my head for years. I just didn’t have the right language for them yet. What changed was not the speed at which the ideas appeared, but the distance between having an idea and being able to capture it, explore it and turn it into something another person could understand. That process has become dramatically faster and, mostly, it happens through conversation.
The interview
Earlier this year I wrote about the first version of this process in The Interview. Instead of asking AI to write something for me, I asked it to interview me. I would provide an initial idea, story or observation, then let it ask questions until we had explored the thing properly. Only after that would we try to turn it into an article.
That addressed a problem I had struggled with for most of my life. I am dyslexic, and my difficulty with writing has never been a shortage of ideas. My thoughts tend to arrive as networks rather than sentences. One observation connects to another, which reminds me of something that happened years ago, which suddenly changes the meaning of the thing I started with.
The traditional writing process asks you to flatten that network into a line. For me, it has always felt like trying to catch smoke. The thought exists, but converting it into a neat sequence of words often changes it before I have properly understood it myself. Talking is different because I can follow the thought while it is moving.
The interview gave me somewhere to put those thoughts while they were still alive. I could explore a connection, follow a tangent and come back later while retaining the shape of the original idea. That was already useful for articles, but the book pushed it much further.
Removing the middle step
I have become increasingly uncomfortable with the way people describe AI and creativity because many of the explanations focus on the wrong thing. They talk about AI generating text, but that is the least interesting part. The interesting part is how it changes the relationship between thought and output.
Traditionally, writing has involved a fairly large production step in the middle. You have an idea, convert it into sentences and paragraphs, then produce enough material to stand back and decide whether it says what you meant. That is where a lot of ideas disappear, not because they were bad but because the cost of getting them into a useful form was too high.
AI has not replaced the thinking for me. It has reduced the friction involved in preserving the thinking while it is still developing. I can dictate an observation, challenge it and change direction without worrying yet about whether it sounds finished.
That distinction matters because AI is very good at finishing things, sometimes too good. Give it half an idea and it will happily turn that half into a complete argument. The writing might be polished before the thinking has caught up. A conclusion that sounds convincing is not necessarily one that has survived examination.
The process is therefore deliberately messy at the beginning. An ordinary observation might lead somewhere unexpected: from attention to hunting, evolutionary behaviour, organisational systems or the way some people acquire authority. Some threads become chapters, some become paragraphs and some disappear entirely. That is fine. The final work does not need to contain all of the exploration. In fact, it probably shouldn’t.
A system for thinking
As the amount of material grew, another problem appeared. I could generate more thinking than I could comfortably hold in my head. I can hold a surprising amount of interconnected information when deeply focused, but interruptions are expensive. Once the chain breaks, rebuilding it takes effort.
The process gradually became externalised. There are conversations, branches, research notes, summaries, chapter drafts and checkpoints. When something reaches a useful state, I preserve it. If I have already written a story elsewhere and spent time getting the wording right, that becomes the source. I do not want a later summary quietly rewriting it into something neater and less true.
The style guide evolved in the same way. It began as a list of preferences: British spelling, first person, reflective tone. Over time, it became a record of where the process tends to go wrong.
AI has a habit of breaking prose into lots of short dramatic paragraphs because it assumes that creates pace. Often it doesn’t. It can make thoughtful writing feel designed for social media. It also tends to remove uncertainty. If there is an awkward or unresolved part of a story, it wants to tidy it away because tidy writing feels more confident. But uncertainty is sometimes the important bit, and a cleaner story is not always a truer one.
Every time I found myself correcting the same kind of mistake, I tried to move that correction earlier. That is something I have always done in businesses and software. If you keep fixing the same problem manually, eventually you should probably fix the system creating it. I am no longer really working inside a document. I am working inside a system for thinking that produces documents.
Seeing the shape before the words
The other half of the process is something I only noticed recently. Despite being dyslexic, I can scan text very quickly. Not read it in the traditional sense, but scan it. I can look at a page and sense that something is wrong before I know exactly why. There might be too many paragraphs, an abrupt change in density, an artificial rhythm or a conclusion arriving too neatly.
I am often seeing the architecture before I am processing the individual sentences. That makes sense when I consider how I tend to think. I usually look for patterns and relationships rather than individual pieces of information.
Because I already know where the ideas came from, I don’t need to discover the argument from the finished text. I can look at the structure first, then slow down and examine the areas that feel wrong. The question is whether its shape still reflects the thought that produced it, not simply whether the passage reads well.
That creates an interesting loop. My dictation is getting faster, AI can turn the raw material into something visible almost immediately, and I can assess its shape much faster than I could listen to somebody read it back. It is less about writing faster than reducing the distance between thinking, capturing and reviewing.
What still takes 45 years
The obvious danger is that fluent output can create the illusion of understanding. AI is very good at producing things that look finished. That is incredibly useful when the underlying thinking is strong, but it is also why judgement matters more than ever.
Over the six days I spent working on the book, I repeatedly had to say that something was too neat, that two ideas were not actually the same thing, that an anecdote did not prove the conclusion drawn from it, or that something I said had been made to sound cleverer but less true. Occasionally I also said, ‘Yes, that is exactly it.’ That is the collaboration.
The machine can hold more text than I can comfortably keep active in my head. It can research, compare versions, find repeated patterns and help organise a huge amount of material. What it cannot do is know which observations matter. That still comes from making things, failing, watching systems work and fail, building companies and relationships, surviving difficult periods and spending years trying to understand why people behave the way they do. AI has not accelerated any of that.
I am not claiming that somebody can reproduce this process by copying a clever prompt and spending a weekend talking to a chatbot. Without something worth capturing, all you have done is automate the production of words. The interesting thing is the opposite.
For most of my life, the bottleneck has not been generating thoughts. It has been getting those thoughts into a form where they can be examined, challenged and shared. The interview reduced that gap. Dictation reduced it further. External memory, branching conversations, research and structured editing reduced it again.
The result happens to be a 45,000-word book written in six days.
But I have not found a way to think 45 years of thoughts in six days. I have found a way to capture them much closer to the speed at which they arrive.
The book that came out of this process is The Hunting Party. You can read the launch post or buy it on Amazon.
You can explore the book and the ideas behind it at thehuntingparty.net.
