Running many agents to hide latency harms human focus and flow. Task management tools do not solve this. Faster models will reduce latency over time. A fast model can handle human communication while a powerful model works. Chat interfaces constrain human input and output. Screen recording with audio lets humans explore and refer to on-screen content naturally. Shelley now supports show-and-tell screen recording with audio and transcription. DOM recording tracks in-page interactions and correlates them with audio feedback. Agents can process rambling, unstructured human input effectively. HTML artifacts with structured information and recording buttons improve human-agent interaction. Live-streaming partial feedback to agents can hide latency without breaking concentration. Fewer agents with better human feedback loops can restore attention and productivity.
What it unlocks
Humans can now provide richer, more natural feedback to agents via screen or DOM recordings with audio.
Sources