Google Web AI Summit demo: 98% of a coding agent's tokens processed on-device with local Gemma
jason_mayes · x · 2026-11-03
AgentGlass recaps three numbers from Google's Web AI Summit: in a hybrid demo, 98% of a coding agent's tokens were processed on the developer's own device — a cloud model planned while local Gemma models built — and in a second demo an agent found and fixed a failing test with Wi-Fi off and no API key. An airline benchmark showed 77% of websites let agents complete some tasks but fail at others, so founders buying agent access need proof on real customer tasks. For security leads, browser or harness code can enforce action limits that hold even if the model is fully prompt-injected. Demo details are in the thread's replies.
More from coding & agent
- 10,000 AI agents burned his API credits; 500 agents in 50 waves worked — Beneficial-Cow-7408 · 2026-11-03
- Walrus MCP Server connects AI assistants to decentralized storage — modelcontextprotocol · 2026-11-03
- News Volume Trends MCP queries news volume growth for any topic — modelcontextprotocol · 2026-11-03
- Pen ships a canvas plugin for Codex, adding design powers to ChatGPT — paw_lean · 2026-11-03
- Elicit engineers all use Ramp's Inspect — a credit card company builds a top coding agent harness — jungofthewon · 2026-11-03
- VibeArmory launches macOS private beta: monitoring and firewalls for AI coding agents — agf2007 · 2026-11-03