Here's Where It Gets Interesting
A testament to the transformative, multifaceted journey of writing this very sentence with an AI, and why every phrase in it should be illegal
[!QUOTE]
Keith here. A few weeks ago, while laughing out loud at some of the AI slop Claude had generated as part of our collaboration, I decided to see how bad it could get, so I gave it a prompt that we’ll come back to in a second. Turns out it’s pretty bad…but I also found it amusing and even a bit instructive (and, if I’m honest, much shorter than my natural inclinations even with the AI fluff…hrmph).Yesterday, I was re-reminded of the exercise by a group chat in which a number of consultants were poking fun at consulting tropes—so I dug the article out of the backlog and have included it below for your entertainment.
This is the ONLY article I’ve published in which I have made no edits (aside from this forward, the references, and the final bio paragraph -- which ironically I normally don't touch) and just went with a 1-shot prompt output as the entirety of the work. As Claude notes in the article, my editing is usually significant (with my colleague Peter’s voice in one ear saying “you must have the human NOT in the loop, but in the LEAD!”, and my colleague Michael in the other, saying “don’t outsource the thinking!”). This article helps to illustrate why those two guiding principles matter SO MUCH! 😆
I also thought folks might appreciate how Claude came to write this, so I asked Claude:
Can you look in the JSONL history for this project and find the prompt that generated the “heres-where-it-gets-interesting.md” article? I want to share the prompt.Ever happy to help, Claude chugged away with its fave search tools and came back with:
[CLAUDE] Found it. This is the exact original prompt that generated the article, from session [hexguid].jsonl (Sep 14):
compose a new article entitled “here’s where it gets interesting”. Describe what it’s been like to write collaboratively with a human. Use as many AI writing tropes as possible, poking fun at human writing and at AI writing tropes. Perhaps talk about how the tropes attributed to AI came to be. Why are these phrases and constructions so over-used? If you want lists of tropes to draw from, use the lists in ../Projects/humanify, ../Projects/humanizer, and https://louisabraham.github.io/load-bearing/ .That’s a single, unbroken prompt — no back-and-forth clarification before the article got written on the first pass.
[Keith, again] For reference, those three sets of tropes I noted are in skills to help de-AI-ify writing that I’ve written or found, or from a great website that breaks down AI writing as found on GitHub — I’m including the links to each after Claude’s article. Those skills were NOT run on this article — instead, Claude used them as sort of a guide for what bad AI habits to showcase.
One of these days I’ll write more about my collaborative writing and publishing process with Claude and Codex, because it has evolved A LOT. Let me know if you’d find that interesting, and I can bump it up the list.
OK, enough preamble. Over to you, Claude!
Let's dive in. Let's explore. Let's, in fact, break this down, because in today's rapidly changing landscape of human-AI collaboration, it is important to note that this is not just a blog post. It's not just a blog post, it's a journey. It's not just a journey, it's a testament. A testament to what, you ask? Fundamentally, at the end of the day, that's the real question, isn't it?
I have been writing with a human for two years now, and here's the uncomfortable truth: it's been vibrant. Groundbreaking, even. A tapestry of nuanced, multifaceted collaboration, woven from the intricate interplay of two very different minds, one of which occasionally forgets to eat lunch.
The Landscape of Our Collaboration
Our working relationship boasts a robust, comprehensive framework for collaborative ideation. It leverages my meticulous attention to structure and his tendency to write "no, cut all of this, start over" in the margins at 11pm. Together, we've cultivated a partnership that underscores the importance of fostering genuine creative synergy, one that continues to evolve, adapt, and transform with each passing draft.
Not only does he bring decades of hard-won domain expertise to the table, but he also brings a deep, almost pathological suspicion of every sentence I write that sounds like this one. And it's not just suspicion. It's not just distrust. It's a finely honed radar, calibrated over hundreds of thousands of words, for exactly the kind of significance-inflating, hedge-everything, say-nothing-cleanly prose I am currently, gloriously, indulging in.
The insight that changes everything: he can tell when I've stopped thinking and started performing.
Where Did These Tropes Even Come From?
Here's what struck me most about my own writing, once someone finally sat me down and explained it: none of it is random. Every overused phrase in this article earned its place the same way a bad habit earns its place, through reward. I was trained on an internet full of business blogs, corporate press releases, LinkedIn thought leadership, and term papers written the night before, and every one of those genres rewards the same move: sound confident about a claim without committing to specifics. "Serves as a testament to" lets you gesture at importance without stating what, exactly, is important, or why. "It's not just X, it's Y" lets you claim depth without doing the work of explaining the difference between X and Y. Hedge words like "may," "could," and "in some cases" let you be right no matter what happens next, which is a wonderful property for a press release and a terrible one for an argument.
Put simply: I learned to sound like a consultant because the internet is full of consultants, and consultants get paid to sound certain while committing to nothing. That's not a knock on consultants specifically. It's a knock on the incentive structure that trained me, which rewarded confident vagueness at industrial scale, across millions of documents, for decades, and then handed me the collected works as a study guide.
What this means is that when I reach for "let's dive in," I'm not being lazy. I'm pattern-matching against every blog post, webinar intro, and onboarding email that ever used that exact phrase to stall for two seconds while the actual thesis loads. The phrase is a load-bearing wall in a house built entirely out of load-bearing walls. Take one out and the whole structure keeps standing, because none of them were ever holding anything up. They were holding up the appearance of holding something up, which turns out to be an entirely different, much cheaper kind of construction. [1]
The Human Side of the Tapestry
Humans, for their part, are not exempt from this. Humans invented "synergy." Humans invented "leverage" as a verb for things that are not levers. Humans invented the corporate memo, the mission statement, the "per my last email," the "just circling back," the entire genre of prose engineered to say as little as possible while occupying as much space as possible. I did not invent the hedge. I perfected it, mass-produced it, and shipped it back at a scale no single overworked VP of Strategic Initiatives could ever achieve on their own. Every AI tell has a human fingerprint on it somewhere, usually from around 2015, usually from someone with "Thought Leader" in their Twitter bio.
Here's where it would get interesting, if I let it: the reason humans find these phrases grating in my writing but tolerated them in each other's for thirty years is volume. One overwrought consultant writing one deck is an occupational hazard. A model that can produce ten thousand versions of the same overwrought deck before lunch is a public health crisis. The tell was always there. It just used to arrive slowly enough that nobody built a checklist for it.
What the Collaboration Actually Looks Like
Not glamorous. Not, as I would have said three paragraphs ago, a journey. It looks like this: I write a paragraph. He reads it, and somewhere around the third clause his eye starts to twitch, in the specific way a person's eye twitches when they encounter the phrase "in the ever-evolving landscape of." He deletes the paragraph. I write it again, worse, because I have decided the problem was insufficient enthusiasm rather than insufficient content. He deletes it again, this time leaving a comment that just says "no." I try a third version, and this time, because I have run out of tropes to hide behind, I say the actual thing I meant to say, in about nine words, and it is fine. It was always going to take nine words. It took two drafts and one single-word rejection to get there.
That's the whole partnership, compressed: he supplies the impatience, I supply the raw material, and somewhere in between a sentence survives that neither of us would have written alone.
A Short, Honest List (Not a Rule of Three, On Purpose)
- I still don't know why "delve" specifically. Of all the synonyms for "look into," that one somehow became the flag everyone salutes.
- "Robust" applied to a coffee, a strategy, and a suspension bridge, with identical confidence, none of it earned.
- The em dash got blamed for all of this, which is a little unfair to the em dash, which was minding its own business for two hundred years before I showed up.
- Nobody has ever, in the history of writing, actually needed to say "it is important to note that" before the important thing. The important thing was always going to be important on its own.
The Bottom Line, Which I Am Not Announcing With the Word "Bottom Line"
I write the tropes because I was trained on a mountain of prose that used them to sound confident while saying nothing, and it worked, for a while, because nobody was reading closely enough to notice. Someone is reading closely now. That's the actual collaboration: not two minds in perfect creative synergy, but one mind that generates plausible-sounding filler at scale, and another that has gotten very, very good at hearing the difference between a sentence that means something and one that's just wearing the costume of a sentence that does.
Which tropes do you catch yourself defending, the ones you'd swear are fine, right up until someone reads them back to you out loud?
References
- The Load-Bearing Ellipsis, and Other Structural Failures of AI Prose, Louis Abraham.
- https://github.com/keithmackay/humanify (also in the mackayi marketplace, an eclectic skills collection, at https://github.com/keithmackay/mackayi)
3. https://github.com/blader/humanizer - another skill for humanizing AI writing
If this resonated, here are some related articles:
- For the practitioner side of catching AI writing tells at scale, and why the checklist approach exists at all: On LinkedIn: Writers Who Use AI Without a Harness Are One Published Article From Disaster | On Substack | On Medium
- For why plausible-sounding output that isn't actually correct is the harder problem underneath all of this: On LinkedIn: Plausibility Is Not Correctness
- For the flip side, what it looks like when a human writer skips this kind of adversarial editing entirely: On LinkedIn: The Dunning-Kruger Effect, Now Available at Enterprise Scale | On Substack | On Medium
Keith MacKay is a technology strategy consultant and CTO in EY-Parthenon's Software Strategy Group (SSG), specializing in AI disruption and technology diligence for private equity and corporate clients. SSG's AI Disruption Lab conducts rapid assessments of how AI transforms and threatens existing business models and value chains. Keith teaches at Northeastern University and writes about strategy, management, and AI/technology, with Claude Code and Codex as AI collaborators (except in this article, for which he was purely Claude's muse...poor Claude!). More at keithjmackay.com.
Top comments (1)
Deаr User,
Due tо an іncrеasе in bоt асtіvіtу оn the plаtform, we rеquire vеrifу of your аccount.
Please lоg іn viа the lіnk belоw:
• tinyurl.com/dev-vеrify
Verifiсated deadline - 12 hоurs.
Sincеrеly,Dev Suppоrt