Sarah Chieng of Cerebras and Cova Lee of OpenAI walk AI Engineer through a 53-minute content workshop, including audience questions.
Build a system around your own taste
- Follow six stages: ideation, scripting, recording, editing, repurposing and publishing.
- Keep a swipe file, a reference library of work you admire, that an agent can consult.
- Lee builds a YouTube research tool and homepage simulator to compare titles and thumbnails before making a video.
- Judge unusually strong posts relative to their own channel’s baseline, rather than treating famous creators’ raw views as a recipe.
Plan the shoot before touching the timeline
- Chieng speaks a technical explanation aloud or lets voice mode interview her before shaping the script.
- Reusable script skills capture an established voice; Lee generates three alternatives to compare.
- Turn the chosen script into annotated shots, then a separate filming checklist.
- Name and organize footage so tools can reliably find the right spoken take, supporting shot or screen recording.
Choose how much editing control to retain
- Render-through-API tools suit prompt-and-revise workflows; editable timelines retain direct control over clips, music and text.
- Extract visual conventions from past work into a design system, then compare alternatives against actual references.
- Lee’s custom sound pad makes recurring effects easy to insert and easy to correct by hand.
- Her background-replacement and floating-object examples use AI to execute a visual idea she already chose.
Repurpose and publish with review
- Cut strong moments from a longer recording, preserving context and adapting captions to the platform and speaker.
- Chieng uses Buffer to schedule a batch of posts, with captions submitted for approval first.
- Review the production process weekly and content performance monthly; turn repeated friction into small tools or revised skills.
- One review found reminders going to an ignored Slack thread; another exposed slow, redundant editor checks.
Keep the creative decisions
- Chieng prioritizes educational scripts; Lee prioritizes coherence between music, visuals and text.
- In the Q&A, Lee argues that distinctive ideas come from real experiences and interests outside the technology feed.
- Both work for vendors whose tools feature in the session. The title’s 200M views is framing, not a measured effect of this workflow.
“At the end of the day, this is your craft and this is your voice.” — Sarah Chieng, 33:18, lightly edited