AI is getting too good for public

hello internet people 👋,

This is the first newsletter I’ve ever written, which is slightly embarrassing considering I’ve made hundreds of posts online, but here we are.

We just crossed 50k on Instagram too, which is insane. Thank you if you came from there.

Ironically, I’m trying to spend less time on my phone at the exact moment more people are finding me through one.

Anyway, AI is getting cheaper, faster, and harder to ignore.

Fable 5 is back and released from prison.

Available only until July 7th on your subscription, so go out and build your dream while it’s cheap...

Sonnet 5 dropped.

I barely got to touch it because Fable 5 came back immediately after and stole my attention. My read: if Sonnet was already your daily driver, this probably will not feel insane. Same price, more intelligence is still a win.

Nano Banana 2 Lite is for speed (7x faster & 2x cheaper)

Good for testing ad angles and visual directions. Still, I would not trust it for final hero creative. Find the angle there, then take it to ChatGPT Image 2 or Nano Banana Pro.

NotebookLM made doom scrolling for education (thanks to Gemini Omni Flash).

You can now turn learning into short vertical videos.

Brain2Qwerty v2 is very cool.

Meta is getting closer to solving telepathy by translating brain activity into typed text. Also, it is still early research.

Fable 5 is back, but it is not the same Fable 5 we saw the first time (prison changed it).

They clearly had to add guardrails before bringing it back. Nonetheless, it is still really good.

The funniest part is that most people online are using it to one-shot prompt games, instead of doing some complex, groundbreaking tasks.

The annoying part tho: it is only included on subscription plans until July 7.

After that, you are paying through API credits.

BUT here is the bigger story.

Fable 5 raised the standard for what frontier intelligence can feel like, then immediately introduced the new problem: the best models might be too dangerous to release to the public.

From now on, I see the AI intelligence being gated.

Some will sit behind API pricing. Some will go only to approved partners. Some will get nerfed before the public gets them.

The best intelligence may not live in the normal chat app anymore. It may live behind usage caps, approvals, enterprise access, API credits, and a very annoying billing page.

Sonnet 5 dropped this week too.

From the stats, it's better. More intelligence, same vibes, still cheaper than the heavy models.

My honest read: if you were already using Sonnet every day, you probably won't feel some insane difference immediately.

That doesn't make it bad.

A lot of useful AI updates are actually quite boring.

Google quietly had a strong week.

→ Nano Banana 2 Lite is 7x faster and 2x cheaper, which makes the use case pretty obvious: fast creative testing.

Watch the demo here

I would not use it for final visuals. But for ad angles, rough concepts, and early visual directions, it makes sense.

Test 10 directions cheap. Find the one that has legs. Then move the winner into ChatGPT Image 2 or Nano Banana Pro.

→ NotebookLM also added Short Video Overviews (thanks to Gemini Omni Flash).

Doom scrolling, but for a good cause.

You can turn complex sources (documents, YT videos, etc.) into reels that explain a concept. It is rolling out now to Google AI Ultra and Pro subscribers on mobile and web, with free users coming soon.

Watch the preview demo here

Meta's Brain2Qwerty v2 research is very cool.

The simple version: Meta is training AI to decode intended typed text from brain activity.

Goodbye, Neuralink?

Right now, when you want to use a computer, you need some kind of input: keyboard, mouse, touchscreen, voice, maybe glasses soon.

Brain2Qwerty points at a different future.

Instead of typing the word, the system reads patterns from your brain while you think about typing, then predicts the text you meant to write.

It’s still in research, but the direction is insane.

We went from typing, to tapping, to voice, to AI agents clicking around for us.

The next interface might be intent.

Imagine wearing something like Meta Ray-Bans and having your thoughts turn into text, commands, or actions without touching a keyboard.

That sounds impossible right now, but so did talking to a computer like a person 10 years ago.

Very early. Very weird. But definitely worth paying attention to

I built a Claude Code skill called /analyze.

Paste a TikTok, Reel, or Short URL and it will:

It runs locally and is basically free to use.

It uses yt-dlp to download the video and Groq Whisper to transcribe the audio.

I built it because I got tired of looking for random websites every time I wanted to get a transcript of a reel.

So I made my own.

Instructions here ⬇️

https://github.com/artemnovitckii/analyze-skill

One idea I keep coming back to:

Find the person who already did the due diligence.

If someone you trust has spent 20 hours testing tools, breaking workflows, comparing setups, and figuring out what is worth using, you do not need to start from zero.

You can speedrun the learning.

That is the point of my free Skool.

I post weekly discoveries there, and we do live Q&A so you can ask me anything about AI tools, content systems, workflows, automations, or whatever you are trying to build.

Join the FREE Skool here

Also, a lot of people asked how I make carousels like these:

So I made a full YouTube tutorial:

Watch the carousel tutorial

I am trying not to only consume AI stuff.

I think a lot of good ideas come from leaving the niche for a second.

Currently reading:

What is one thing AI could do for you, but you still do not let it touch?

For me it is handwritten notes.

I know I could build some insane second brain setup, tag every thought, summarize every page, and pretend I am becoming limitless.

BUT I still like writing in a notebook, I truly believe that it helps me retain more information.

I want to know what’s yours.

See you next week,