Veed’s Sabba Keynejad: 99% of Companies Shouldn’t Build Their Own AI Model (41 min)
ai-driven-innovation-economy
ai-in-art-music-creation
ai-in-everyday-life
- Release date: 2026-06-25
- Listen on Spotify: Open episode
- Episode description:
Sabba Keynejad, co-founder and CEO of Veed.io, says 99% of companies shouldn't build their own AI model. Veed did it anyway.Sabba joins Wade Foster and Dan Slagen (SVP Marketing at Zapier) on Agents of Scale for a grounded conversation about the state of video AI. Why it's still 2-3 years behind LLMs. Why Veed built Fabric 1.0 despite every reason not to. And what most companies get wrong about AI video, YouTube, and creativity itself.In this episode:- Why video AI is 2-3 years behind LLMs- Why 99% of companies shouldn't build their own model- The "backed into a corner" Fabric 1.0 story- Why AI slop is a creation problem, not a technology problem- The 0.5-second rule for video retention- Why SaaS companies are sleeping on YouTube- How AI lowers the barrier to creativity without replacing creatorsSubscribe for more conversations with the operators and builders turning AI from experiment into infrastructure.
Summary
- 🎥 Video AI Lags LLMs: High latency and cost keep video generation 2-3 years behind text models, limiting broad adoption despite strong use in marketing.
- 🛠️ Owning the Model Stack: Veed built Fabric after third-party dependency issues, achieving 60x lower cost and 7x speed for talking-head content.
- 🧠 Elaborate Workflows Win: Savvy creators chain multiple specialized models and apply sharp creative judgment to avoid ‘AI slop’ and produce standout results.
- 🚀 Daily Video as Default: Companies that publish short, hook-driven clips daily see massive reach; AI lowers the barrier to consistent creation.
- 🔮 AI Video Becomes Just Video: In 12-24 months, AI clips will be indistinguishable from traditional video and treated as another creative medium.
Insights
- Why is video AI still 2-3 years behind large language models in terms of adoption and reliability?
- Time: 2:14 – 2:45
- Answer: Latency, high generation costs, and the need for elaborate multi-model workflows are holding back widespread use, even as marketing teams and Hollywood experiment successfully with talking-head content.
- How will the line between ‘AI video’ and ‘real video’ disappear, and what does that mean for creators?
- Time: 6:43 – 7:19
- Answer: Within 1-2 years, AI-generated clips will simply be another video medium like animation or stop-motion, rewarding those who master constraints and novel use cases rather than chasing photorealism alone.
- When does owning your AI stack become essential rather than a costly distraction for product companies?