In order to prevent AI from hallucinating, you must narrow its seemingly infinite possibilities into something you can control. We are orchestrating the work so that AI can do it for us.
Our responsibility has shifted. It is less about pushing pixels or writing every line of code manually. It is more about structuring the problem, setting boundaries, defining expectations, and giving AI a clear playing field. So that AI can dance inside the garden we design.
AI will skip and contradict itself if you write a page-long or two pages of prompts. You have to limit the prompts and instruct it to wait for follow-up prompts. Think of it like an outline. You have to go line-by-line, prompt-by-prompt to get to the conclusion and ultimately the output you seek from AI.
You must have system and project instructions to govern the ‘action’, ‘narrative’ and ‘output’.
So, in a sense, chain prompt. I wrote about how to structure it all with an example in my most recent newsletter.
It’s crazy how many people treat AI like a magic box instead of a powerful tool that requires direction. By putting in the effort to create the right detailed prompts, you’re not just getting better results; you're unlocking a whole new level of creativity.
100% resonate with using AI as an intellectual sparring partner. If you treat it as a co-collaborator and peer, it will reciprocate. You have to deliver rich and deep context input to get the same depth of output. Excellent breakdown.
People miss this context because the cultural buzzword is 'Agent AI'. The value of an LLM is not in its agency, but its ability to act as an accelerated for the user to personalize faster.
Another great article—as Dan did with his 'Human 3.0' prompt. If you haven't read or used it, put it on your capture list now. There is an art form—a developmental form—to this prompting. Funny, I call my prompt builder "Frank." I had "Frank" build this query using JSON (I don't know it) to generate an image of a Ninja character out of one of the stories I'm writing, and there she was rising out of a book as the writer's pen touches the page.
There are many forms of psychology and many forms of martial arts, so setting a base level requires some thought. In the case of Dan, he's using Ken Wilber's model.
Over the last few years, I've developed a personal process that I call The Code, and I will be writing about it. It synthesizes 20 or 30 different sources—from Freud, Adler, and Jung to modern versions like Mark Solms. A mindset, framework, and procedure.
A lot of this comes down to structure for both the competitive and non-competitive markets. We can bring in military strategists and start using structures that are already there. The bottom line is there are frameworks—psychological or procedural—that can be plugged into a prompt to add structure. The key, too, is the gift of sparring partner feedback; tell the AI to ask you questions.
The frustrations around hallucinations also point to something important: the need for guardrails. The Anthropic/Claude website has written well on this, and the concept mirrors martial arts training. You don't get stronger by fighting without boundaries; you get stronger by sparring inside a defined structure. Prompts work the same way. A well-designed boundary gives the model something firm to press against.
Much of my work draws from Major John Boyd's, nicknamed '40-Second Boyd' for his Top Gun teacher skill of getting on a student's tail in 40 seconds, OODA loop—the classic decision cycle illustrated by the four-part iterative diagram —Observe, Orient, Decide, Act—an iterative system that produces clarity, feedback, and resilience. At the center of it all is self-concept. The "synthesis" was flushed out with AI, but it did not create my concepts; instead, it became a teacher in a box, as Steve Jobs called the big intention of the PC.
I resonate with this approach of using AI to learn why something works and how to replicate the structure but with your own ideas. I think it's helpful to go through this brainstorming process as well to see if there are any gaps in your own thinking and I agree, once you have a good outline, I think it's important to do the writing yourself so you keep your unique voice. Thanks for all the prompt guides!
In order to prevent AI from hallucinating, you must narrow its seemingly infinite possibilities into something you can control. We are orchestrating the work so that AI can do it for us.
Our responsibility has shifted. It is less about pushing pixels or writing every line of code manually. It is more about structuring the problem, setting boundaries, defining expectations, and giving AI a clear playing field. So that AI can dance inside the garden we design.
Nice summary
Nice summary
Mastering AI prompt engineering may be the highest leverage skill available since the advent of the internet.
Great read.
Signed up for the waitlist for Eden. Looking forward to see what you and your cofounders have been cooking.
AI will skip and contradict itself if you write a page-long or two pages of prompts. You have to limit the prompts and instruct it to wait for follow-up prompts. Think of it like an outline. You have to go line-by-line, prompt-by-prompt to get to the conclusion and ultimately the output you seek from AI.
You must have system and project instructions to govern the ‘action’, ‘narrative’ and ‘output’.
So, in a sense, chain prompt. I wrote about how to structure it all with an example in my most recent newsletter.
No way Dan gave that prompt for free after I paid for it in "Productize Yourself" 😂
FS ahahah
Slop creates slop. If you can only make slop, AI can only give you slop.
Knowing how to use AI is crucial nowadays. Whether you want to accept it or not.
But there are boundaries that you must impose while using it in order to maintain your control and humanity over your work.
I wrote something about it, in fact. I’ll drop it here from some curious minds.
https://open.substack.com/pub/tidecultureandhistory/p/nah-grammarly-im-good?r=5rxtn3&utm_campaign=post&utm_medium=web&showWelcomeOnShare=false
I think Eden's gonna be cool and all but I still use Kortex and I really don't want to leave it, Dan. 😅
I'm literally writing my next letter on Kortex right now.
Why am I unable to read or view anything in this? Is there any issue?
It’s crazy how many people treat AI like a magic box instead of a powerful tool that requires direction. By putting in the effort to create the right detailed prompts, you’re not just getting better results; you're unlocking a whole new level of creativity.
100% resonate with using AI as an intellectual sparring partner. If you treat it as a co-collaborator and peer, it will reciprocate. You have to deliver rich and deep context input to get the same depth of output. Excellent breakdown.
People miss this context because the cultural buzzword is 'Agent AI'. The value of an LLM is not in its agency, but its ability to act as an accelerated for the user to personalize faster.
I already follow a similar approach. Thanks for the detailed guide.
Spot on. Excellent
Its just a great professional approach of prompting ;
thanks because of the complete explanations
TO ME, AI IS SIMPLY AN ALFRED TO MY BRUCE 🔥
Another great article—as Dan did with his 'Human 3.0' prompt. If you haven't read or used it, put it on your capture list now. There is an art form—a developmental form—to this prompting. Funny, I call my prompt builder "Frank." I had "Frank" build this query using JSON (I don't know it) to generate an image of a Ninja character out of one of the stories I'm writing, and there she was rising out of a book as the writer's pen touches the page.
There are many forms of psychology and many forms of martial arts, so setting a base level requires some thought. In the case of Dan, he's using Ken Wilber's model.
Over the last few years, I've developed a personal process that I call The Code, and I will be writing about it. It synthesizes 20 or 30 different sources—from Freud, Adler, and Jung to modern versions like Mark Solms. A mindset, framework, and procedure.
A lot of this comes down to structure for both the competitive and non-competitive markets. We can bring in military strategists and start using structures that are already there. The bottom line is there are frameworks—psychological or procedural—that can be plugged into a prompt to add structure. The key, too, is the gift of sparring partner feedback; tell the AI to ask you questions.
The frustrations around hallucinations also point to something important: the need for guardrails. The Anthropic/Claude website has written well on this, and the concept mirrors martial arts training. You don't get stronger by fighting without boundaries; you get stronger by sparring inside a defined structure. Prompts work the same way. A well-designed boundary gives the model something firm to press against.
Much of my work draws from Major John Boyd's, nicknamed '40-Second Boyd' for his Top Gun teacher skill of getting on a student's tail in 40 seconds, OODA loop—the classic decision cycle illustrated by the four-part iterative diagram —Observe, Orient, Decide, Act—an iterative system that produces clarity, feedback, and resilience. At the center of it all is self-concept. The "synthesis" was flushed out with AI, but it did not create my concepts; instead, it became a teacher in a box, as Steve Jobs called the big intention of the PC.
I resonate with this approach of using AI to learn why something works and how to replicate the structure but with your own ideas. I think it's helpful to go through this brainstorming process as well to see if there are any gaps in your own thinking and I agree, once you have a good outline, I think it's important to do the writing yourself so you keep your unique voice. Thanks for all the prompt guides!