I’ve spent years building SaaS products, AI agents, automation systems, and video-generation tools.
Here are 12 lessons I’ve learned about building software, getting customers, and turning ideas into businesses. 🧵
I'm training a small decision-making model, similar to JEV, but specialized for my own use case.
The goal is simple:
- Specialized in one niche
- Fast inference
- Runs entirely on CPU
- No expensive GPU required
I'm going to document the whole journey, from dataset preparation and training to evaluation and inference.
Would you be interested in following this experiment?
@atoiansk what if it sounds like AI, so we must ban photos|videos generated by AI because it looks like AI "he makes money from selling AI photos", what hurt the platform is automated AI replies not posts written with the help of AI
X could basically fix the so called AI Bot problem tomorrow if they wanted to and mostly have already.
So in a nutshell this means the megalomaniac is just shutting down people (he does not like)
For automated AI replies that use the public API. X already did a major version of this.
On February 23, 2026, they changed the API so that on self-serve tiers (Free, Basic, Pro, Pay-per-use), the POST /2/posts endpoint only accepts a reply if the original author has “summoned” the account (by @mentioning it or quoting one of its posts).
Otherwise the call is rejected. They explicitly said this was to stop LLM-generated reply spam. That single change killed the common pattern of bots scanning popular posts and auto-replying under them.
What X could still do tomorrow
They fully control the API, so they could:
Disable the reply endpoint entirely for most tiers
Further tighten the “summon” requirement
Revoke or rate-limit specific app keys
Add stronger automated detection of AI-style replies
Enterprise access is currently exempt from the February restriction, but that is also under X’s control.
Important caveats
Manual replies are unaffected. Someone can still copy-paste an AI-generated reply into the app or website.
The API change only stops programmatic replies. (bots)
Official/privileged accounts (including @grok when summoned) operate outside the normal self-serve limits.
Non-API methods (browser automation, scripts that drive the website) are already against the rules and can lead to permanent suspension.
In short: for the large class of automated AI reply bots that rely on the API, X can (and already largely has) shut them down by restricting replies through the API.
The remaining problem is mostly manual copy-paste or accounts that somehow still have broader access.
So tell me, how is this list that is promoted by the court jester and sycophants any more than a hate list?
Please educate me........
@atoiansk what if it sounds like AI, so we must ban photos|videos generated by AI because it looks like AI "he makes money from selling AI photos", what hurt the platform is automated AI replies not posts written with the help of AI
Woke up today wondering why I had lost around 20 followers.
Then I found out.
An indie hacker influencer had published a list of accounts allegedly using AI to reply. My account was on it, and people started blocking everyone on that list.
The funny part? I don’t use AI to write my replies. I was apparently put on that list because I disagreed with one of his takes on science.
Imagine having that much influence over people’s reputations without having to prove anything.
Since when did disagreeing with someone become a reason to publicly blacklist them?
This is exactly the kind of behavior that will makes X a worse place for open discussion.
@atoiansk It’s not written with AI, second what your point?
Everyone use AI as assistant in a way, even X has improve post with grok. So I didn’t get what is the relation between getting help from AI to improve a post and make automatic replies with AI.
Some influencer-worshipping NPC are calling me out for using AI to write posts.
Let’s get one thing straight: I’ve never claimed I don’t use AI as an assistant. I use it to write. I don’t use it to fake conversations or automate replies.
There’s a reason @ThreadPilots homepage explicitly says we don’t automate Reddit.
Could I add automated replies and make a lot more money, like some influencers do? Absolutely.
But I’d rather build something useful than another spam machine.
Using AI isn’t the problem. Automating replies is.
@timwiesnerer Sorry but this is a shit tool, no way you can detect ai, even openai wrote an article about it even when they leave a fingerprint in the generated text and they still can’t detect its ai
Here is the link to prove that it’s written by human
https://t.co/D8Dh0oFwWJ
@georgevichdev@levelsio Sorry but this is so embarrassing and humiliating, he’s the one who should apologize in public and swear an oath to never do it again.