Agentic Video is Real and Here's What You Should Try
Use AI video first where showing beats telling, existing recordings need editing, or repeated explanations can become updateable assets.
- Start with an existing meeting recording and ask AI to find key moments and make a short highlight reel.
- For a new video, provide the concept, source material, and length; ask an AI coding tool to plan, visualize, and edit it.
- Build a recurring pipeline only after testing a one-off video, such as a monthly update from current metrics or documentation.
The examples and three-level starting plan are practical, but there is no hands-on tool walkthrough. · Best for: Knowledge workers exploring internal communications, training, customer education, or thought leadership video.
The speaker argues that recent AI models and connected tools make video editing and creation accessible for more workplace tasks. The most promising uses include editing existing footage, explaining visual or changing information, and maintaining videos through automated pipelines. AI does not replace skilled video professionals; it makes previously uneconomical video projects more feasible.
What changed in AI video3:47
- Editing can be directed with natural-language requests, such as finding good takes, trimming filler, placing B-roll, or making a first cut.
- Models can search video or transcripts for particular moments across large recordings; connectors can provide meetings, calls, documents, and other source material.
- Agents can record software demos using computer-use capabilities; Claude Code can create animations with React and Remotion.
- Generated footage, audio, avatars, and dubbing add options, but generated clips remain short, results can feel uncanny, and editing may require multiple passes.
- Video features are appearing in existing tools and in general-purpose coding harnesses; cost can rise with advanced generation and editing.
When video is a better format10:31
- Use video when movement or pacing helps show a UI, workflow, or changing numbers better than text or a still image.
- Consider audiences unlikely to read a document, explanations repeated often, and information needed in multiple languages.
- Turn existing recordings into shorter edits, or use video when conveying conviction and emotion matters.
Internal workplace uses16:33
- Create concept videos for proposed products, processes, customer journeys, or store layouts; use prototypes, walkthroughs, floor plans, or mock-ups as anchor visuals.
- Make short versions of important meetings when the debate or exchange itself matters, rather than relying only on a written summary.
- Curate customer calls into themes and video clips for decision-makers; create executive pre-reads and role-based onboarding.
- Build videos that update themselves as source material changes; animate metrics, and turn spoken handoffs into useful records.
External video uses23:50
- Repurpose existing talks, webinars, podcast appearances, or audio as short-form clips.
- Try long-form thought leadership: ramble to camera and have AI edit it, or use AI to shape a clear script in your voice before recording.
- Generate personalized account videos from a base video; keep customer education current with a recurring update pipeline.
Three levels to get started27:52
- Level 1: Ask AI to edit an existing recording, such as creating a meeting highlight reel.
- Level 2: Generate something new, such as an animated chart from two quarters of spreadsheet data or a product concept video.
- Level 3: Build a recurring pipeline, such as a weekly agent-generated update from metrics, tickets, or shipped features.
- The goal is to use video for work that previously did not justify the production effort, not to replace expert video creators.
Step by step
- Choose a use case where video could beat text: for example, a key meeting, repeated explanation, workflow, or changing metrics.
- Start with the simplest level: give an AI tool an existing recording and ask it to identify important moments and create a short edit.
- For a new video, state the concept, audience, approximate length, and available source assets; collaborate with the AI on visuals and structure.
- Review and refine the result; expect that editing may take multiple passes and generated footage may not suit every purpose.
- If updates are recurring, build a pipeline with a trigger, source-material search, script refresh, and video rendering; test a one-off first.
Tools & resources mentioned
- Loom — record a walkthrough or ramble-to-camera video.
- Claude Code and Codex — manage video tasks and, in the speaker's examples, build or run video workflows.
- Remotion — render React-based animations for videos.
- Descript Agent Underlord — conversational video-editing capability mentioned.
- Adobe Firefly AI assistant and Paper Edit for Premiere — Adobe video-assistance tools mentioned.
- CapCut — an example of a tool for short-form vertical video.
- OpenAI image generation APIs — add image-generation capability to a Claude Code workflow.
- Google Agentic Video Understanding — video-understanding capability discussed; the speaker cites Google’s reported efficiency and accuracy gains.
- Kling, Gemini Omni, Veo, and Runway — video-generation models discussed for creating footage and audio.
Notes are AI-generated from the video's transcript and may miss nuance.