Summary
Ekaterina Deyneka, founder and CEO of Reelful, presents the company’s agentic video editing system at the AI Engineer conference, opening with the observation that most people record far more personal footage than they ever edit and share — and positioning automated editing as the solution. Unlike generative video tools that work from a blank canvas, Reelful focuses exclusively on real user-supplied footage: identifying the best moments, cutting pauses and failed takes, generating captions, music, voiceover, and B-roll, then assembling a polished clip without manual intervention.
The infrastructure architecture closely resembles an agentic app builder: a sandboxed remote machine spins up on demand, running an agent equipped with craft-encoded skills — cut rules, font pair guidelines, B-roll generation parameters — that encode Reelful’s editorial taste directly into the agent’s capabilities. The composition layer uses Remotion, an open-source framework for building videos as React code, which plays to agents’ strength at code generation. A verification layer checks the final composition for errors and triggers agent iteration before rendering if problems are found.
Deyneka identifies the core challenge as harder than blank-canvas generation: the agent must make judgment calls about which real moments matter, handle messy or incomplete footage, and produce output indistinguishable from human-edited work. She describes a mobile-first UX strategy — allowing users to submit footage while commuting or exercising — and a conversational interface designed to reduce the prompting burden for consumers unfamiliar with AI workflows. A live demo of a completed edit is shown during the talk.
📺 Source: AI Engineer · Published August 18, 2026
🏷️ Format: Keynote Launch
