Hermes Agent, NotebookLM & LiveKit Founders on the AI Agent Race | TWiAI 17

TL;DR

• 1

  Hermes Agent went from internal tool to top open model app: Jeffrey Canel says Hermes Agent is now roughly number one on OpenRouter, passed OpenClaw in popularity, and won users with polish, reliability, and a product built by people who used it obsessively themselves.

• 2

  NotebookLM is shifting from separate tools to one integrated agent: Steven Berlin Johnson says Google just merged NotebookLM's chat, artifact creation, and deep research into a single agent that remembers your sources, research history, and prior outputs, so it can answer questions like "what am I missing?"

• 3

  LiveKit is the infrastructure behind major voice AI products: Russ d'Sa says LiveKit powers voice experiences for Spotify, Tesla support and robotaxis, xAI's Grok Voice, Salesforce Agentforce Voice, and SAP's Joule, positioning it as a lower layer in the stack rather than a Zendesk or Salesforce competitor.

• 4

  The graduation boos came from a broken social contract, not just anti-tech vibes: The panel argues students were told for decades that college and white-collar knowledge work were the path to status and stability, then watched AI threaten entry-level jobs just as they entered a brutal labor market.

• 5

  Apple's Siri reboot still has a core UX problem: Russ likes the improved voice and Gemini partnership but says Siri still feels transactional and unclear about what it can actually do, while Steven is more excited that Apple is finally building a new AI-native app surface at all.

• 6

  Token maxing works for top performers and fails when it becomes expensive laziness: Jeffrey says one NUS researcher spends around $1 million annualized on tokens and more than earns it back, but warns that giving everyone unlimited AI budgets can just produce the same work at twice the cost.