5 Rules for Better AI Writing

August 27, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
August 26, 2026

MAIN STORY

Five Rules for AI Writing

When one of finance's biggest voices dropped an op-ed in The Wall Street Journal critiquing the Treasury Secretary and current monetary policy, the discourse that followed wasn't about monetary policy at all — it was about the fact that the piece was very, very clearly written by AI. Then, instead of a sheepish apology, both the famous financier and the Journal's opinion editor said yes, of course he used AI, which quietly settles the disclosure question and opens a much better one. If AI writing is now a fact of life, how do we make it actually good?

Let the Bond Market Speak
"That isn't a crisis. It is an invoice."
Druckenmiller has legendary status among finance folks — among other achievements, a hedge fund with zero losing years between 1981 and 2010 — and the piece cuts to the quick of macroeconomic policy right now, so from a substance perspective it matters. Except the substance was never what people were talking about, because the op-ed is replete with just about every egregious AI-ism that exists today, starting with it's-not-this-it's-that. AI detectors like Pangram put the odds it was AI generated at 100%, and plenty of readers were not happy: Jesse Livermore called presenting AI slop as your own writing plagiarism, full stop.

Of Course I Used AI
"I write everything using AI now… same reason I use a calculator"
Rather than apologizing, Druckenmiller's response was essentially bro, of course I used AI — there's a reason he moved from an English major to an economics major, he's not embarrassed by it, and at 73 he's kind of proud of using it. Opinion editor Paul Gigot backed him up completely, saying AI is a fact of modern life and the only questions that matter for the Journal are whether what it publishes reflects an author's original argument and whether the author has the standing and credibility to make it. Clear as day, the Journal's op-ed position is that what matters is the sincerity of the ideas, not the unique human flavor of the words used to express them.

The Speechwriter Precedent
Nobody ever staged a controversy over the job "speechwriter"
A lot of folks pointed out this wasn't all that different from how leading figures have done op-eds for quite some time. Joe Weisenthal noted that prior to AI there were plenty of high-profile people publishing op-eds under their own name that were entirely written by some underling, without much controversy — and that it's always been faintly surprising that speechwriting is completely normal and expected. Validating it, journalist Sharon Goldman wrote that from 2015 to 2020 she wrote dozens of op-eds for top executives on the back of a single conversation and some notes, never receiving or wanting a byline.

AI Is Just a Tool
"If he puts his name behind it, it's his opinion"
The largest camp treated the whole thing as a category error. Jason Calacanis argued that publishing an AI-written piece without disclosure in the first sentence is unforgivable for a public figure because readers can't tell what's the author's and what's the magic black box's; his All In co-host Chamath Palihapitiya bit back that this is a dumb new form of virtue signaling, since nobody discloses every article they've ever read before spouting off on X. Aria Deniz put the constructive version of it best: if AI helps an investor take decades of experience, judgment and pattern recognition and articulate those thoughts more clearly, that's a good thing, because the value is in the thinking.

Contempt for the Format
The problem isn't that they're wrong — it's what it expresses
For others this was a much more nuanced conversation, and the sharpest version came from Dirty Texas Hedge: what Druckenmiller and Gigot say is 100% true descriptively, and the problem is that it represents an expression of contempt for the format of an op-ed and for its audience. Brent Donnelly reduced the whole debate to a single question — is writing art, or is it a functional communication tool like code — and concluded there's no right answer, it's a personal thing. David McMahon landed in the middle, with no doubt the piece reflected Druckenmiller's actual thinking but plenty of concern that too many people are using AI as a substitute for serious thinking.

AI Is Great for the Middle
"Beginning needs a human… end needs to be human too"
Deirdre Bosa offered the framework that ends up organizing most of this: the beginning needs a human deciding what the idea is and what you're actually trying to say, the middle is where AI researches, organizes, drafts, rewrites, tightens and pokes holes, and the end needs a human again asking whether this is any good, whether it's true, whether you buy it. Her read is that Druckenmiller didn't do the end part as well as he could have. The thought she was reaching for originally came from Balaji, back in June of 2025: AI doesn't do it end-to-end, it does it middle-to-middle, and the new bottlenecks are prompting and verifying.

The Foundation: AI Writing Is a Fact of Life
Writing dominates enterprise usage across almost every department
Anyone needing evidence can look at OpenAI's recently released enterprise data on how people use ChatGPT: the preponderance of usage across almost every department is writing, and not just for comms — recruiting, marketing, customer support, legal, sales, policy, you name it. AI writing is here and it's here to stay. But that doesn't mean all AI writing is the same, which is where the rules start.

Rule One: Different Types of Writing, Different Rules
An email is not a strategy memo is not an op-ed
These different types of writing are trying to achieve different things, and the way anyone engages with AI around them will consequently be different. Which means "can or should AI write this" is, in most cases, simply the wrong question.

Rule Two: The Purity Test Goes, the Quality Test Doesn't
The purity phase goes the way of the dodo. Quality doesn't.
The folks saying this conversation will look quaint in a couple of years are frankly probably right. But if anything, the increased ease of the inputs of writing increases the burden on the outputs even more — people starting to accept that AI was used does not mean they'll accept bad writing as a consequence. Joe Weisenthal's version of the prediction is that refusing to let LLMs write for you will become as incomprehensible as refusing Google Maps in a new city; Paul Graham's counter is that it will be more like choosing to run and lift weights even though machines can do both, and therefore prestigious enough that more people will claim to write for themselves than actually do.

Rule Three: Quality Is Effort, and Effort Is Obvious
Perceived laziness in the prose reads as laziness in the argument
For most people, quality corresponds with effort or the perception of effort, and its absence tends to be pretty obvious. A lot of the negative response to the op-ed was that the AI-isms were so glaring and so fixable that it seemed like nobody cared enough to change the syntax on a couple of sentences — and if that wasn't worth the time, why would anyone believe much thought went into the point underneath. The perception of a lack of effort in the writing becomes an expression of a lack of effort in the construction of the argument.

Rule Four: Longer Is Not Better
Usually the opposite is true, and enterprise readers feel it
The reason the phrase work slop exists is that a lot of the models used over the last couple of years have been much better at saying a lot than at saying the right thing succinctly. Great writers have always known this — Blaise Pascal, in the 17th century, apologized in a letter for having made it longer than usual because he had not had time to make it shorter. The first wave of AI writing was defined by endlessly long, eye-bleeding documents. The next generation will not be.

Rule Five: Writing Is Thinking
Not a guarantee of outsourced thinking — a constant risk
Deciding what your thesis is, what the supporting arguments are, how the narrative gets constructed, even weighing the impact of a turn of phrase — the point of these exercises is ultimately not the words that end up on the page but the substance of the argument running underneath them. Using AI for writing does not a priori mean outsourcing your thinking, but it's the risk that needs to hover in your field of view the entire time, because it is so easy to fall into. The tension shows up in something as ordinary as ideating an episode concept: handing the whole thing to Claude with a title and two of the rules would have been much faster than sitting in Notion arguing with yourself over whether it's five rules or six — and for this particular context, the slower manual approach was the right one.

Crib Sheet: Emails
Safe for AI, and increasingly overkill
For the vast majority of emails no one has ever judged the quality of the prose, and emails focus on such small, concise units of thought that the intellectual-outsourcing risk barely registers. Standard responses to inbound requests are a fine use. But for interactive email — the kind of thing you'd otherwise do in Slack — dictating with something like Wispr Flow is likely faster than prompting.

Crib Sheet: Meeting Notes
Very safe, because it's compression rather than writing
An AI that sat through 60 minutes and holds the full transcript will be very good at summarizing what everyone said, and much worse than you at explaining the most important thing anyone said. The fix is to outsource the bulk of the summarization and then come in over the top with the one or two things that actually mattered.

Crib Sheet: Internal Strategy Memos
Sneakily bad, because strategy is exactly where writing is thinking
It looks like a great AI use case since prose quality barely matters, but unless you're hyper-precise about the strategy being communicated and the things it is meant not to communicate, it's very easy for AI to go off the rails, add its own material, or produce generic advice drawn from training data rather than your organization's actual context. Split the interaction into two separate levels: use AI to support the thinking, bulleting and outlining, then treat the final writing as an entirely separate process. If the outline is tight and carries the effort, letting AI write the actual words is usually fine.

Crib Sheet: Social Media Copy
Mixed — and the platform doesn't reward the copy anyway
AI does a fine job on a lot of social copy, and does better the shorter the medium: an old-school single-line post on X is easier to land than a full LinkedIn essay, which simply has more room for AI-isms to shine through. The bigger problem is that social media doesn't reward pithy writing so much as engagement — it wants posters commenting on other people's posts, not just automating more content into the maw. Invest heavily up front in tuning the voice, and hold realistic expectations about the ceiling.

Crib Sheet: Marketing Copy
Annoyingly bad, precisely where you most want it to be good
Marketing is the area where the most egregious AI-isms turn up, maybe because it's a domain that wants uniqueness and distinction rather than sameness. There's also a specific failure mode: give the model a correction like "stop presuming so much about the reader" and it turns the correction itself into copy, returning a headline that says "We assume nothing about the reader." No reliable fix has surfaced yet beyond feeding it extensive on-brand examples — but as a rapid iterator it's excellent, and asking for 20 taglines is a great way to spark your own.

Crib Sheet: Op-Eds and Convincing Essays
Do the thinking first. This is the five-paragraph essay, baby.
The whole point of an op-ed is to convince someone of something, and when readers can instantly tell it was low effort they switch off — if you couldn't be bothered to make your argument convincingly, why should anyone care about the argument. That's what happened here: perceived laziness reduced the perceived strength of the underlying case. Get extremely clear on the thesis and the supporting points before turning it over to AI, then complete the effort sandwich by going back through and removing the incredibly obvious AI-isms.