I'm tired of the AI tone | Sagiv Ofek

Skip to main contentsagiv ofekblogangelopen sourceAll posts2026I'm tired of the AI toneQuoted $1M for AI code review. Built it for free.The Rise of the Backwards Deployed EngineerDistribution is the New Engineering2025SaaS is the new mallSpecialist is Dead. Long Live the Generalist!Everyone Sounds the Same2024Knowledge is a Commodity2023How to plan a team offsiteHow to choose your startup investors2022Startup ideas for 2022A confession of a FAANG hiring managerBack in the arena2021Crypto BillboardDear recruiter - a canned response to a canned emailTime is (your only) moneyStartup ideas for 20212020How to become an ETH2 validator?How much is enough?Enjoy the journeyGoodbye Medium2018The missing Dictionary in Solidity2015Hackathon As Part Of Your Company DNAGo Native — Calling C++ From NodeJS2014Permutations — Ruby Style Recursion I'm tired of the AI toneSeptember 20, 2026 · 5 min readI'm tired of the AI tone.
Not AI writing itself. AI is incredibly useful. I've probably used ChatGPT more this week than I'd like to admit.
I'm talking about the weird, increasingly recognizable way everything on the internet now sounds like it was written by the same extremely articulate 27-year-old who has never had a bad day.
You know the tone.
Everything is a wedge.
Every idea is not just X, but Y.
Every paragraph contains an em dash — like this — because apparently the comma has been deprecated.
Everything is "a powerful shift."
Something doesn't just solve a problem. It "unlocks a new way to think about the problem."
A product isn't useful. It "sits at the intersection of X and Y."
A company doesn't build something. It "reimagines how we..."
And every post somehow ends with:
The future is already here.
Thanks, Claude.
The AI vocabulary
There are certain words that have become suspicious.
"Wedge" is probably the biggest one.
Every startup now has a wedge.
"We're starting with X as our wedge into Y."
Nobody talks like this in real life.
I've been in startups for a long time. I've sat in hundreds of product meetings. I've talked to founders, engineers, customers, and investors.
I have almost never heard someone say:
«"This is our wedge."»
But put the same person in front of ChatGPT and suddenly they're building a "land-and-expand wedge into the enterprise."
Other suspicious words include:

unlock
leverage
ecosystem
paradigm
transformative
seamless
robust
thoughtfully
empower
foster
nuanced
landscape
intersection
fundamentally
increasingly
reimagine
accelerate
journey

None of these words are bad.
That's what makes it worse.
They're perfectly normal words that AI has managed to make feel like corporate spam.
And then there are the em dashes
Why are there so many fucking em dashes?
Seriously.
AI loves an em dash.
It uses them to connect thoughts that a normal human would separate into two sentences.
Or a comma.
Or just leave disconnected because humans don't actually write with perfect transitions all the time.
AI writing has this strange obsession with making every sentence flow beautifully into the next one.
Real people don't do that.
Sometimes we start a sentence badly.
Sometimes we use parentheses.
Sometimes we repeat ourselves.
Sometimes we say "and" four times.
Sometimes we write:
«"I don't know. Maybe I'm wrong."»
And move on.
That's writing.
The perfect structure
AI also loves structure.
You can usually recognize an AI-generated post from about 50 feet away.
It starts with a provocative sentence.
Then:
Here's the thing:
Then three or five perfectly balanced points.
Then a counterargument.
Then a thoughtful conclusion.
Then a sentence about how "this isn't about X. It's about Y."
Then the inspirational ending.
It's like every LinkedIn post has gone through the same screenwriting course.
And God forbid you make a list.
AI will give you exactly five bullets.
Not four.
Not six.
Five.
Because five feels comprehensive.
The fake vulnerability
My favorite is the AI version of vulnerability.
"I used to think X.
I was wrong.
After spending years working on Y, I've come to realize Z."
That's not how people talk.
That's how a machine imagines someone talks after analyzing 40,000 LinkedIn posts.
Real vulnerability is messier.
"I completely screwed this up."
"I thought this was going to work. It didn't."
"I don't really know what I'm doing here."
"I changed my mind."
"I have no idea if this is right."
Those sentences are interesting because they're actually human.
The problem isn't that AI writes well
Ironically, the writing is often technically better.
That's the problem.
AI removes all the little imperfections that tell you a human wrote something.
The weird phrasing.
The unnecessary joke.
The sentence that goes on too long.
The opinion that doesn't quite fit the rest of the argument.
The typo.
The "I don't know how to explain this, but..."
Those things give writing a personality.
AI optimizes them away.
And when everyone uses the same model, everyone optimizes toward the same mean.
We are slowly converging on one universal Internet Voice™.
Polite.
Clear.
Structured.
Nuanced.
Optimistic.
Slightly corporate.
Occasionally profound.
And completely fucking interchangeable.
Please sound like yourself
I'm not arguing that we should stop using AI to write.
Quite the opposite.
Use it.
Have it rewrite your terrible first draft.
Have it fix your grammar.
Have it make your argument clearer.
Have it turn your 2,000-word ramble into something people can actually read.
But then put some of yourself back into it.
Delete the em dashes.
Remove "unlock."
Kill the "not X, but Y."
Delete "wedge" unless you are actually talking about a wedge.
Say what you mean.
Use shorter sentences.
Have an opinion.
Be slightly weird.
Don't make every paragraph perfectly symmetrical.
And for the love of God, occasionally write a sentence that sounds like something you would actually say out loud.
AI is getting incredibly good at sounding intelligent.
The competitive advantage now might be sounding human.
Post draft is ready, do you want me to make it shorter?Tags:aiwritingOlder postQuoted $1M for AI code review. Built it for free.CodeGithubStack OverflowSocialTwitterLinkedinInstagramMoreBlogEmail

The author expresses dissatisfaction with the increasingly standardized and artificial tone generated by artificial intelligence, arguing that this homogenized style erodes authentic human communication. The critique targets specific linguistic patterns that characterize contemporary AI-generated writing, suggesting that these patterns create a pervasive, impersonal voice on the internet.

The author identifies several common linguistic markers favored by AI that contribute to this synthetic sound. These include the frequent use of certain buzzwords such as "wedge," "unlock," "leverage," "ecosystem," "paradigm," "transformative," "seamless," and "reimagine." While not inherently negative, the author notes that these words are optimized by the AI to sound like corporate boilerplate, lacking the genuine articulation found in real-world discourse. Furthermore, the overuse of em dashes is highlighted, as the AI tends to use them to create overly smooth, continuous transitions that contrast with the more organic, sometimes disjointed, flow of natural human writing.

Moreover, the author analyzes the structural conventions of AI writing, noting that the AI favors a highly predictable scaffolding. This structure typically involves starting with a provocative statement, following with perfectly balanced points, presenting a counterargument, offering a thoughtful conclusion, and concluding with an inspirational remark. The AI excels at creating balanced lists, often defaulting to exactly five points because this structure is perceived as comprehensive, a tendency that contrasts with the naturally variable structure of human thought and expression.

A particular focus is placed on the nature of vulnerability when expressed by machines. The author finds the AI’s handling of vulnerability—such as admitting a change of mind—to be inauthentic. Genuine human vulnerability is messier, incorporating self-doubt, mistakes, and uncertainty, whereas the AI simulates this vulnerability through perfectly structured reflections that lack genuine imperfection.

The core problem, according to the author, is that the AI optimizes away the minor imperfections that provide writing with a distinctive human personality. These imperfections include awkward phrasing, unnecessary tangents, unusual punctuation, and the subtle deviations that occur when a human formulates an idea. By eliminating these variations, the AI pushes all generated text toward a singular, optimized "Internet Voice"—one that is polite, clear, structured, nuanced, optimistic, and slightly corporate, resulting in content that is interchangeable.

The author does not advocate for abandoning AI tools, but rather for a critical approach to their use. The recommendation is to leverage AI for mechanical tasks, such as refining grammar, improving clarity, and editing drafts, but to consciously reintroduce human elements into the final product. This means actively removing the AI's polished cadence, using shorter sentences, injecting personal opinions, allowing for asymmetrical paragraph structures, and deliberately writing sentences that sound authentically spoken. The ultimate competitive advantage in the current landscape is achieving a tone that sounds genuinely human, rather than merely intelligent.