The show where AI goes from hype to implementation. Real workflows. On screen. Every episode is one you can steal.
Singapore
Joined April 2026
- Tweets72
- Following15
- Followers44
- Likes35
ChatGPT in a tab can't run a command or open a browser. Manus and Codex can, because each agent gets its own virtual computer.
@SherryYanJiang's plain explanation of what makes browser automation work.
No guessing who might sponsor. @SherryYanJiang had Manus pull companies that already sponsored NVIDIA GTC and past AI Engineer events, 500 in total, then filter for growth, BD, devrel and startup founders.
Her reason for those titles: they reply.
The outreach message was a few lines @SherryYanJiang wrote herself. AI did none of the writing. It did the sending, to hundreds of LinkedIn contacts.
On personalisation, her rule from the receiving end: get the name and the company right.
Client sites used to go: scope, design approvals, build, iterations, then changes after launch. @jigyasha555 now gets the design commented on and fully approved while it's built, so it ships with the changes already made.
The prompt that created Kate, an AI influencer for the dictation app Voibe, was written by Claude. Ayush Chaturvedi asked it how to use Fastlane.ai, then pasted its prompt into the influencer form. His rule: wherever you have to prompt, get Claude to prompt for you.
How Kate was made, from first prompt to first post: youtu.be/jwHAu8Fvg2Y
Unable to scroll TikTok from India, where it's banned, Ayush Chaturvedi lets a feed do it: about 100 proven formats a day. He swipes to pick one and asks for his app, Voibe, in the copy. Back comes Kate, his AI influencer, with the hook, then a demo of the app.
From warmed-up US account to scheduled post, the full episode: youtu.be/jwHAu8Fvg2Y
Account farms, as Ayush Chaturvedi describes them: five to 15 accounts, a hook tested on one and copied across the rest once it works, then paid ads behind it. His own setup is one test account, with no paid ads yet.
Why he's starting with one account, in the full episode: youtu.be/jwHAu8Fvg2Y
Most marketers go after AI citations and ignore share of voice, says @jigyasha555: how often the brand comes up, positive or negative, and how it's described. She measures both, prompt by prompt, before and after a page goes live.
The part of her workflow @jigyasha555 dropped entirely: building design options for website pages, one iteration at a time. She now looks at ten options and picks one.
Vibe-coded sites look perfect on day zero. @jigyasha555 says they're missing two things: LLM readability, meaning how easily the content can be extracted, and the brand guidelines that set the tone of voice.
Why he let AI near confidential deal documents at all: shippingwithai.substack.com/…