AI & Automation Briefing - October 23, 2026
1Password Lets AI Agents Log In on Your Behalf Without Accessing Your Credentials
1Password launched a product called 1Password for Agents that allows AI agents to authenticate into user accounts without ever seeing the actual password. The system works through a three-way handshake: when an agent attempts to log in, it sends a unique digital key to 1Password to verify its identity, the user receives an alert and manually approves the request, and 1Password then fills in the credentials on the agent's behalf. Supported agents at launch include Claude Duet from Anthropic, OpenAI's Dots and ChatGPT Work, Grok Bot, Meta's Muse, and agents from AI startup Town. The launch coincides with a report from Axios that U.S. banks are detecting agentic activity inside customer accounts and struggling to determine whether that activity is user-authorized. Banks have flagged credential sharing with agents as a significant security concern, with Chase advising customers not to share usernames and passwords with any third party.
Anthropic Launches Claude Duet, an Agent Built for Professionals and Small Businesses
Anthropic has released Claude Duet, an AI agent targeting professionals and small businesses rather than personal use cases. Unlike personal agents, Duet does not handle errands or show live browser activity. Instead, it builds daily briefings, creates dashboards, and suggests actions based on user data. Duet runs tasks in the background on its own virtual machine and requests user approval before sensitive steps, including purchases made through Stripe Link or Shopify Shop Pay. It will not trade securities or move money between accounts. A new classifier checks each message to block exposure of credit card numbers and government ID. Built on Claude Opus 5.5, Duet connects to over 6,000 integrations through the Claude Marketplace. It is available on iOS, Android, and Claude.ai for Pro and Max subscribers in research preview, with access through a waitlist.
Simple RAG Beats Complex Agentic Memory Systems Across Major Benchmarks
A new paper introduces RAW-RAG, a straightforward retrieval-augmented generation approach that uses dense retrieval and reranking directly over raw conversation history. Tested across four benchmarks including LongMemEval and MemoryAgentBench, RAW-RAG ranked first among compared memory systems in six of seven settings, outperforming the strongest agentic memory system by up to 5.53 points. The researchers found that broader coverage of relevant evidence matters more than sophisticated memory transformations, and that converting raw history into structured memory formats can actually cause accuracy loss. The findings position naive RAG as a necessary baseline before investing in more complex agentic memory architectures.