The Hallway Track
8 tracked signals on tool-use.
Meta releases Muse Glimmer, a 30B open-weights agentic model under Apache 2.0 license
“Muse Glimmer achieves strong success rates on full-task benchmarks including DeepSearch QA, MCP-Atlas, 𝛕-Bench and SWE-Bench, which measure its ability to work within scaffolds, write and debug code, and resolve multi-turn requests from start to finish.”
Google launches Gemini 3.5 Flash, its most capable model yet, surpassing 3.1 Pro on benchmarks
“3.5 Flash, the most capable model ever that we've shipped.”
A Microsoft product designer built a personal AI pet that uses agentic tool use to browse the web, post to Discord, and commit to GitHub.
“all the interfaces you use have agents running behind them, and eventually those will be the things that you have to design for”
Hugging Face expands EVA-Bench to 3 domains, 121 tools, and 213 scenarios for agent evaluation.
llm-gemini 0.33 adds Gemini 3.7 Flash support with reasoning traces and server-side tools
LangChain's managed deep agents now support custom tool integration via Python and TypeScript decorators
LangChain defines an AI agent as an LLM looping through decide, act, reason, repeat.
“That loop, decide, act, reason, repeat. That's what makes an agent.”
Simon Willison released datasette-agent-edit, a base plugin implementing core agentic text-editing tools for Datasette Agent.
“Agentic editing of text is a little tricky to get right.”