Google I/O Afterparty: The Future of Human-AI Collaboration, From Veo to Mariner

Three Google Labs leads — Thomas Iljic (Veo/Whisk/Flow), Jaclyn Konzelmann (Project Mariner), Simon Tokumine (NotebookLM) — reveal how generative video, computer-use agents, and AI notebooks are converging into new content formats and commerce behaviors.

• Project Mariner shifted from taking over the foreground browser to running 10 parallel tasks simultaneously on background VMs after user feedback demanded their browser back.
• Mariner’s companion Chrome extension reads all open tabs to pass live user context into VM-executed tasks, bridging local browser state and background agents.
• Thomas Iljic argues video generation, simulation, and games are converging into a single world-building paradigm where creators set the stage and others shoot or interact inside it.
• Google Labs teams were repeatedly too early — some projects paused and are only now viable as model capability and cost curves caught up.
• NotebookLM’s viral audio overviews were an accident: the team expected a niche RAG workspace tool and was unprepared for the scale, spending the first months keeping TPUs from melting.
• Simon Tokumine sees NotebookLM expanding from audio overviews into context-adaptive formats (comic books, mind maps, short films) tailored to a single user’s project or learning goal.
• Jaclyn Konzelmann predicts agents will act as a universal cart across e-commerce sites, shifting competitive advantage from checkout UX to pure product quality.
• Thomas Iljic believes the hard unsolved problem is not model R&D but the abstraction layer on top — how users define voice, character mannerisms, and audio-visual inputs without long text prompts.

2025-06-03 · Watch on YouTube