The most expensive content marketers make is the content they use once.
Customer interviews kept surfacing the same post-webinar ritual: a marketer spends days clipping highlights, writing the recap blog, drafting the follow-up emails and cutting social posts. The webinar took a month to produce, and its content died within a week.
LLMs and video AI had just made the whole ritual automatable. The question was whether we could ship it as a product, at production quality, before everyone else did.
Prototype-first PM'ing on a brand-new stack.
- Prototyped the pipeline myself: transcript → chaptering → highlight detection → asset generation, to prove output quality before committing an engineering team to an unproven LLM stack.
- Designed for the marketer's calendar, not the demo: outputs mapped to what teams actually publish: clips, recap blogs, email sequences, social posts, quote cards.
- Built evaluation loops for quality: human-graded samples and prompt iteration so "AI-generated" never meant "needs rewriting," the failure mode that kills repurposing tools.
- Shipped 0→1 as the Snackable Content Hub: positioned to extend webinar ROI, reinforcing the Webinar+ pipeline story rather than living as a gimmick feature.
80% less production time, and a durable differentiator.
Marketers went from days of post-event production to minutes of review: one webinar became 40+ on-brand assets, and post-event production time dropped by 80%. The capability became a core differentiation point for Webinar+ and survived the acquisition, Brandlive now ships AI-powered highlights, clips and on-demand content hubs as a centerpiece of its BrandTV offering.
This is the AI-native loop at full speed: I researched the workflow, prototyped the pipeline, validated output quality with real users, then built it with engineering. No spec preceded the proof.