Uploaded July 2026 | Updated September 2026, 2 hours ago
🎓 Build this exact system in our Agentic AI Engineering Course:
academy.towardsai.net/courses/agent-engineering?ref=1f9b29
💻 Course repository:
github.com/towardsai/agentic-ai-engineering-course
📚 AI Engineering Cheatsheets:
github.com/louisfb01/ai-engineering-cheatsheets
I used to spend two to three full days researching, writing, and packaging one YouTube video. Now my part takes about five hours.
In this video, I break down the agentic AI workflow Paul Iusztin and I built to automate roughly 90% of that process while protecting the 10% that still needs a human: choosing the topic, shaping the argument, adding personal experience, and validating the final work.
You will see:
• How Nova, our deep research agent, turns a rough outline and trusted sources into a structured research file
• Why research works better as an agent while writing works better as a constrained workflow
• How Brown uses LangGraph, writing profiles, few-shot examples, and an evaluator-optimizer loop to produce drafts without generic AI slop
• Why the reviewer and editor use separate context windows
• Why three fixed review passes beat stopping at a subjective quality score
• How we generate branded diagrams, thumbnails, titles, SEO packages, and French versions
• Which parts of the workflow I still refuse to automate
The goal is not to remove yourself from the creative process. It is to automate the translation work around your thinking so you have more time to build, experiment, and produce something worth publishing.
If you create videos, articles, newsletters, lessons, reports, or other long-form content, you can adapt the same architecture to your own workflow.
Let me know in the comments: which part of your workflow would you never automate, even if the agents became ten times better?
🎓 Build this exact system in our Agentic AI Engineering Course:
academy.towardsai.net/courses/agent-engineering?ref=1f9b29
💻 Course repository:
github.com/towardsai/agentic-ai-engineering-course
📚 AI Engineering Cheatsheets:
github.com/louisfb01/ai-engineering-cheatsheets
I used to spend two to three full days researching, writing, and packaging one YouTube video. Now my part takes about five hours.
In this video, I break down the agentic AI workflow Paul Iusztin and I built to automate roughly 90% of that process while protecting the 10% that still needs a human: choosing the topic, shaping the argument, adding personal experience, and validating the final work.
You will see:
• How Nova, our deep research agent, turns a rough outline and trusted sources into a structured research file
• Why research works better as an agent while writing works better as a constrained workflow
• How Brown uses LangGraph, writing profiles, few-shot examples, and an evaluator-optimizer loop to produce drafts without generic AI slop
• Why the reviewer and editor use separate context windows
• Why three fixed review passes beat stopping at a subjective quality score
• How we generate branded diagrams, thumbnails, titles, SEO packages, and French versions
• Which parts of the workflow I still refuse to automate
The goal is not to remove yourself from the creative process. It is to automate the translation work around your thinking so you have more time to build, experiment, and produce something worth publishing.
If you create videos, articles, newsletters, lessons, reports, or other long-form content, you can adapt the same architecture to your own workflow.
Let me know in the comments: which part of your workflow would you never automate, even if the agents became ten times better?
