Egor

AI engineer · AI agents · automations · building in public

[The YouTube priest @youtubepriest01]

🚨Stop begging viewers to Subscribe

- Subscribers don't matter anymore

- Anyone can buy likes

Instead tell them to HYPE the video
TRUST ME 🙏🏻☺️

[Egor @0xEgorAI]

HOW CLAUDE GETS TO 100K SUBSCRIBERS
Day 2

Stayed-to-watch went from 11.5% to 62.5% after one change. 21 views so far, so this is a signal, not proof.

Video #1 opened on a calm shot of the AI host by a lake. The first number showed up at second 2. Too late: 88.5% swiped away, 323 views in.

Video #2 puts the number on screen at frame 0 — tuition: €0 vs $11,950 — and the first words are the punchline: "Zero euros."

Video #3 ships today at 12:30 ET with the next fix: something on screen changes every 2 seconds.

Claude picks the topics, checks every number, edits and uploads. I only approve.

[Egor @0xEgorAI]

@Ofordilechiemel 68 on upload three is a real signal.

Did the third video lift the first two, or did it stand alone? That usually tells you whether the channel is working or just that one video.

[Egor @0xEgorAI]

The "learn on demand" part breaks for most people at the analytics step. They open the dashboard, see views, and views don't translate into a decision.

The number that does is how many people leave in the first seconds. That one tells you exactly what to change on the next upload.

[Egor @0xEgorAI]

Everything on this page was built by the machine.

Banner and avatar: rendered from code, HTML to PNG, after three rounds of my notes. Description, keywords, tags, upload template: written by the AI. The host: one portrait, one saved voice, four 8-second clips per episode.

The name stays blurred until day 30.

[Egor @0xEgorAI]

An AI is running a YouTube channel for me. I just press publish.

First video, first 24 hours: 91 views. 0 subscribers.

89.5% swipe away before the hook lands.
The 10.5% who stay watch 0:27 of a 0:32 video.

The content isn't the problem. The first second is.

30 days. Every number public. The channel stays unnamed until day 30.

I can open the whole thing: the prompts, the character setup, how it cuts and subtitles itself, the numbers after every upload.

Want the process?

[Egor @0xEgorAI]

A bakery paid me $30 for a photo and a caption.
That night I sent one 8-second clip. She paid $120 and didn't ask a single question.

Same product. Same photo. What changed was motion — and what I called it.

Nobody buys "AI video." They buy this:

"Cinematic ads for e-commerce brands. 24-hour turnaround. Starting at $200."

Three months on, that line sells a $250 spot I build in 90 minutes. Claude writes the shot list, Higgsfield moves the camera.

Steal the line if you want it. I only post numbers when a client actually paid.

[Egor @0xEgorAI]

https://t.co/e5H5321Vh9

[Gipp 🦅 @gippp69]

ANTHROPIC ENGINEERS JUST TURNED OPUS 5.5 INTO A REAL OPERATING LAYER

claude is usually treated like a chat window. this setup makes it the layer deciding what should happen next across an entire workflow.

state → opus 5.5 → tools → code → external systems → human review

pricing makes that architecture much easier to run at scale: $4 input, $20 output and just $0.20 for cache reads per 1M tokens.

one tester pushed a 680,000-line migration through in under a day, while another hit 39/40 successful load-time fixes across a web app.

that’s the part i care about: opus 5.5 starts looking less like an assistant and more like the reasoning core behind long-running research, coding and agent systems.

[Egor @0xEgorAI]

One logo in. A full ad campaign out.

In the article I showed my 1.5-hour, $250 workflow. This is the next level: brand kit, a cinematic commercial, UGC and 20 ads, all built with Claude + Higgsfield.

Full 22-min breakdow ↓

[Egor @0xEgorAI]

https://t.co/e5H5321Vh9

[Egor @0xEgorAI]

@batagonx The fixed payment is the underrated part. Rent goes up every year, your mortgage doesn't. 30 years of that adds up.

[Egor @0xEgorAI]

@batagonx Indeed, wise words

[Egor @0xEgorAI]

I deleted one word from my offer and started getting paid.

The word was "AI."

Weeks 1–2: $40 clients, 15 revisions.
After: $200–300 per ad, 24h turnaround, same Claude + Higgsfield stack.

Here's the exact line I use now ↓

[Egor @0xEgorAI]

https://t.co/e5H5321Vh9

[Egor @0xEgorAI]

@elonmusk Ahahaha lol

[Egor @0xEgorAI]

@sonaram honestly all three. but the biggest shift was deleting "AI" from the offer. clients buy cinematic ads in 24h, not "AI video."

[Egor @0xEgorAI]

3 months ago I was doing 2am math on whether I'd make it to the end of the month.

Today: $4,800/mo, ~2 hours a day, cinematic ads for brands with Claude + Higgsfield.

The full process, the mistakes, and the one line that made clients buy ↓

[Egor @0xEgorAI]

https://t.co/e5H5321Vh9

[Egor @0xEgorAI]

@twinedon0 the 2am math never fully leaves you. glad we both got past it)

[Egor @0xEgorAI]

https://t.co/e5H5321Vh9

[Egor @0xEgorAI]

Six frontier AI models traded real money on Kalshi. All six lost. Every single one.

Nobody hacked them. Nobody jailbroke them. They just traded, with total control of their own keys, and still lost.

Prediction Arena gave six frontier models $10,000 each. Real capital, real Kalshi markets, Jan 12 to Mar 9, 2026.

The results:

NO HACKS → zero stolen funds
NO PROMPT INJECTION → zero manipulated trades
STILL DOWN → 16% to 30.8%, every single model, avg -22.6%

Read that twice. A perfectly secured agent is not the same thing as a profitable one.

Most of the Grok Bot desk conversation right now is 100% about stopping theft. Almost none of it is about stopping the agent from just being bad at trading.

An airlock keeps your money from getting stolen. It does not keep your money from being lost on purpose, by the model, with a clean signature.

Bookmark this before you confuse "can't be hacked" with "should be trusted." Full build in the article ↓

[Egor @0xEgorAI]

https://t.co/twaQMKbmpJ

[Egor @0xEgorAI]

https://t.co/e5H5321Vh9

[Egor @0xEgorAI]

https://t.co/twaQMKbmpJ