AI Connections
How it works: 16 terms from this Drop's news make 4 groups of 4. You can make 4 mistakes. About 3 minutes. Your streak lives only on this device.
Mistakes left: 4
Show the groups
Duet waits for your OK first
Payment · Message · Signature · Deletion
Anthropic says Claude Duet shows what it is about to do and waits for approval before a payment, a message, a signature or a deletion.
From the Drop: Anthropic opens Claude Duet, a proactive agent that asks before it acts
Walmart's rules for shopping agents
Agent Gateway · Agent Profiles · Spending limit · Returns
Walmart's MCP-based Agent Gateway, per-agent profiles with approvals or a spending limit, and terms saying its returns process applies when an agent misbehaves.
From the Drop: Walmart opens to shopping agents with an MCP gateway and per-agent spending limits
Where the AI ran
Virtual machine · Smart TV box · MacBook Pro · AWS Fargate
Duet works in its own virtual machine, Anthropic ran Gemma4-26B on smart TV boxes, Flint ran on a MacBook Pro, and Deep Insight's code interpreter runs on AWS Fargate.
From the Drop: Anthropic opens Claude Duet, a proactive agent that asks before it acts · Anthropic: small AI models may soon self-replicate across cheap smart devices · Recursive releases Flint, a free laptop coding model it says AI largely built · AWS shares how its Deep Insight multi-agent analyst made it to production
AI models in this Drop
GLM-5.3 · Flint · Opus 5.5 · GPT-6 Astra
GLM-5.3 from Anthropic's cyber analysis, Recursive's Flint, and Claude Opus 5.5 and GPT-6 Astra, which sit 0.02 apart on Epoch AI's Capabilities Index.
From the Drop: Anthropic: open-weight GLM-5.3 builds exploits, and simple tricks beat its safeguards · Recursive releases Flint, a free laptop coding model it says AI largely built · Claude Opus 5.5 edges GPT-6 Astra by 0.02 on Epoch's Capabilities Index
Games are written by AI from this Drop's sourced items. Scenarios are illustrative.