I watch a lot of podcasts and video content. It's one of my primary sources for ideas, frameworks, and insights that make their way into this newsletter and my client work. But turning video content into LinkedIn posts has always been a friction point.
The workflow used to look like this: watch the video, take notes, draft a post, find or create a visual with another tool, edit, publish. For a single post from a 30-minute video, I might spend an hour or more.
So I built something to solve it.
Castifai turns YouTube videos into ready-to-publish content. The workflow is simple:
Paste a YouTube link - the app pulls video information and extracts the transcript
Choose your content type - LinkedIn Post, Newsletter section, Infographic, Summary, Quiz Questions, or Flashcards
Customize - select length (short/normal/long), style (Facts & Stats, Process/Steps, Comparison), image style (Modern, Corporate, Playful, Data Viz, Comic), and output language, and optionally add custom instructions
Generate - the app creates your content, including a visual if you chose Infographic
Share - post directly to LinkedIn or download for other uses
I also added My Channels management feature. You can save your favorite YouTube channels, and the app shows you their latest videos ready to transcribe and transform - no need to hunt for links.
The two infographics in this issue section 1 and 2 - the OpenAI vs Anthropic comparison and the AI in Education overview - were both generated by Castifai using Nano Banana Pro API.
I built Castifai using Base44, an AI-powered no-code platform. Base44 lets you describe what you want in natural language, and it generates a working application - database, authentication, hosting, and all.
The interesting part isn't my initial prompt. It's the process I use when vibe coding like this.
I start with an intuition that this tool needed to exist and a rough idea of what it should do. Then I began experimenting. Base44 has a "Discuss mode" where you can brainstorm with the AI before it writes any code. I used this extensively - planning features, thinking through edge cases, exploring different approaches before committing to implementation.
When it came time to build, I select Claude Opus 4.5 exclusively in the editor - at least for now. It costs 2.5x the credits, but the difference in effectiveness is dramatic. For complex features, the higher capability model saved me significant debugging time.
The real iteration came from users. I shared early versions with friends and contacts on LinkedIn, collected feedback, and kept improving. I even tried using Claude extension in Chrome and OpenAI's Atlas browser to test and suggest UX improvements - but both tools are too early in their development to be useful for this. The most valuable feedback came from actual humans using the actual product.
From first prompt to the current version took days, not weeks or months.
This connects to the third bucket in my AI value framework: Driving Disruption. When anyone can build custom tools to solve their specific problems, the competitive landscape shifts.
Previously, if you had a workflow problem, your options were: find existing software that approximates your need, hire developers to build something custom, or live with the friction. Now there's a fourth option: build exactly what you need yourself, in days.
I'm not a developer. But I now have a production application with user accounts, a content library, channel management, transcript storage, and image generation - because the tools to build it exist.
What workflow friction are you living with that you could solve yourself?
Sign up for free at castifai.com and turn a video into content. I'd love your feedback on what works and what could be better - I have more features in development and real user input shapes what gets built next.