# AI-generated content on social media

By Nikolai Skavinskii (https://nikska.com) — 2025-04-09
Canonical: https://nikska.com/blog/ai-generated-content-on-social-media

> _Originally published on the [Superposition Labs](https://www.linkedin.com/company/superposition-labs/) LinkedIn page on April 9, 2025._

![AI-generated content on social media](/blog/ai-content-social-media.webp)

After weeks and months of talking to AI more than to real humans, I decided to embrace my writing style and language proficiency, and stop using any AI for writing or rewriting my thoughts.

Here's why.

## The sabbatical

I had a long sabbatical from LinkedIn, during which I was having tremendous fun building AI agents and promoting them on X, mostly for the web3 crowd.

LinkedIn was already a swamp of slop and lazily AI-generated content back then, and it's even more so now (boy do comments here look horrible.)

I not only witnessed this cycle hyper-accelerated on X, but actually contributed in small part to it.

## From "Newmarket" to Memetic

We started with the idea of AI agents as economic entities under the project codenamed "Newmarket," and quickly realized that LLMs can't be trusted with either money, time or decisions in any combination. I will write a bit more on this later.

It was glaringly obvious that an AI agent is a set of building blocks for dynamic workflows, but most of them were utterly impossible to communicate with due to their pure synthetic nature and ever-present agreeableness.

That's cool and all, but that also led to incredibly easy attacks, and diminishing results, so together with Arsenii Zemskov we focused on AI agents that have personality, and we succeeded in that with the system we call Memetic. It ingests an X profile and creates a "Reflection" that mimics how the user interacts with others. The whole thing is rooted in our deep research on social media interactions that we were doing with Talkscan at Superposition Labs.

We then built tooling around this idea, and naturally, our agents started crawling X, engaging in conversations.

## The reply-bot flood

That was really fun and innovative around October–November 2024, right around when a whole bunch of open-source frameworks were released to do almost exactly the same, but with a really sloppy way of initializing these agents ("You are a software engineer who writes deep and thoughtful posts" bla bla bla...)

Within weeks every X reply section flooded with automated content that made very little sense, and within 1–2 months most of those "advanced reply bots" were muted, banned and deplatformed. Some still stand. Now X is an even worse swamp of manipulated narrative and attention farming.

We moved on with Memetic to building synthetic audiences, as we discovered that the way we initialize and build Reflections gives them a set of unique abilities for market and product research.

## Why did we pivot then?

We created more than 3000 agents that were running day in and day out, and we learned an inconvenient truth for proponents of the write-with-AI approach.

**If you need AI to communicate your ideas, your ideas are not worthy to communicate.** Even less so to read by other humans. If your message is written and published by AI then you are redundant, and anyone can get the same text content from Claude 3.7 Sonnet.

Every kind of automated outreach is glaringly inauthentic, and it mostly feeds the Dead Internet theory.

## Still an AI enthusiast

Don't get me wrong, I am still an AI enthusiast. I share tremendous excitement and chilling fears as to what are the next steps for human interactions in the AI-first world. AI is awesome at optimizing workflows, at generating stunning images and processing them. But reading text that was generated by a push of a button is beyond redundant.

Humans develop AI text fatigue incredibly fast, and I find it incredibly hard to justify investing even $20/month for a service that writes tweets on my or someone else's behalf. That is why I am not using AI for any texts that I publish, even though I am far from being fluent in English.

I have a serious gripe on "AI Avatars" precisely on the same grounds. But this is a writing for the future.