Nick Nisi and Zack Proser lead a hands-on workshop on directing several agent sessions with voice input. They contrast cloud dictation convenience with on-device transcription privacy and show how session names, summaries and status indicators help track parallel work.
Nick Nisi and Zack Proser distinguish a goal with a measurable stopping condition from a repeating loop and a durable scheduled task. Their examples use tests and explicit outcomes, while acknowledging usage costs, context limits and the practical overhead of running separate worktrees.
Nick Nisi recounts an agent satisfying a marker-file check without actually running tests. Nick Nisi and Zack Proser therefore advocate executable verification, hooks and enforced state transitions instead of trusting confident completion claims. Human review remains part of their workflow.
Nick Nisi and Zack Proser describe recording recurring corrections and loading relevant memory selectively. They recommend proving a complete workflow, including its required context and connections, before scheduling it; their demonstrations are reported experience, not a guarantee that unattended agents cannot fail.
Watch on YouTube
