#49 - On Writing (this newsletter) w/ AI
July 2026
My goal for 2026 is to write around a focused subject: AI + GTM.
There are plenty of reasons I wanted to do this:
I enjoy the writing process
it’s extra motivation to run AI experiments daily, and
it forces me to be consistent and keep on a focused subject.
Most importantly, I really wanted to figure how best to leverage AI for writing.
I see a lot of slop, and don’t like producing slop, but you must be able to drive efficiency without AI destroying the quality - right?
I don’t get a ton of feedback on most of my articles but my recent On Writing (with AI) seemed to strike a nerve.
The comments I got were some variation of “how much of this shit have you been writing with AI” i.e. how long have you been hoodwinking us into thinking this was all your original work?
The Ever-Evolving Process
In December 2025, I set up a system using GPT projects (the equivalent of Claude “skills”).
I uploaded my previous articles to give it an understanding of my style, while also “coding” rules like DO NOT, UNDER ANY CIRCUMSTANCES, USE AN UNNECESSARY EMDASH.
My hope that was by spending a few hours tuning the “brain” I could get as close to a one-shot draft as possible. Below is the system I hoped would work:
Version 1 — AI-first drafting
Step 1: Generate full draft with AI (30 minutes - 1 hour)
Get something complete on the page with a good, detailed prompt. Structurally sound, logically organized.Step 2: Edit and refine (1 hour)
Adjustments on tone, tighten arguments, rework examples.
Total: ~2 hours
This did not work at all. No matter how many times I refined prompts I couldn’t get it to write in my style even after putting in hours of effort to build the brain.
The issue wasn’t that the draft was bad - it was probably, actually, technically very polished and good. I just found myself spending so much time editing that I was effectively rebuilding the article from scratch every time.
No matter what, emdashes, it’s not X but Y, the word “quietly”, and an overall preachy tone that was rather off-putting would keep finding it’s way into the piece.
This, regardless of how much effort I put into codifiying the “brain”, highlighted how LLMs actually work (always pulling towards a reversion to the mean).
Doing it this way actually ended up taking more time than V2 (see below). It also removed a lot of the enjoyment from the writing process.
While prompting can help clarify your thinking in some capacity, it left a lot on the table. Skipping the draft process entirely left me editing thoughts that weren’t really my own.
Version 2 — AI-assisted structuring (what actually works)
Step 1: Use AI to organize thoughts + outline (1 hour)
Dump ideas via voicenote into the “brain” in a rambled, disorganized fashion so that absolutely no context is lost. Have it help with structuring the article by organizing those thoughts around a central point, but does not draft anything. It is very good at doing this and removes a lot of the angst of staring at a blank page…Step 2: Write the article from scratch (6–7 hours)
Build the argument in my own voice. Decide what matters, what doesn’t, and how to say it.Step 3: Light editing + tightening (1-2 hours)
Clean up language, sharpen points, make sure it flows. Good to do this after a few days break.
Total: ~10 hours
In the first version, the work goes into fixing something that never quite feels right. In the second, the work goes into actually writing, which is slower upfront but produces something I’m much more confident in.
Good Content & GTM Success
The reason this matters extends far beyond writing.
In a world where everyone has access to AI (soon to be a commodity), the ability to develop a differentiated point of view is becoming one of the most important drivers of GTM success.
Buyers are inundated with content, most of which says roughly the same thing. The companies that stand out are the ones that can articulate a unique perspective on a market problem and back it up with compelling ideas.
AI is incredibly effective at helping organize information, identify patterns, and apply established best practices. What it cannot do is develop that perspective for you.
By definition, LLMs are derivative. They are trained on what already exists. The most effective GTM strategies, however, rarely come from doing what everyone else is already doing. They come from identifying something others have missed, developing a unique perspective on a market problem, and communicating that idea in a way that resonates with buyers.