AI-generated summary LangChain discusses techniques for agent engineering to manage model context windows and reduce context rot during longer tasks. The open source harness Deep Agents uses middleware to move context out of active windows into files, summaries, or separate agents.
AI-generated summary San Francisco-based Graphon has raised thirty-five million dollars in funding to develop software that helps AI agents store and retrieve information over time. The company states that its software is already used by customers including the San Francisco 49ers and HP.
AI-generated summary SurrealDB CEO Tobie Hitchcock says the real challenge of scaling AI agents is memory rather than inference. Companies like Verizon, Tencent, and Samsung Ads have combined several databases to reduce operational costs and response times.
AI-generated summary Confluent revealed new capabilities for its Streamhouse data architecture at the Current 2026 event to keep business context current for AI agents. Confluent also announced general availability for Lightning Tables on Amazon Web Services during the event.
AI-generated summary Researchers at MIT released VISTA, a new agent harness designed to store images in persistent memory for faster retrieval. The VISTA agent completed interactive visual reasoning puzzles using fewer actions than humans who had not encountered them before.
AI-generated summary MIT researchers have developed a memory system called UQ-DAAAM to help robots identify and correct inaccurate descriptions of objects stored in their memory. The technology evaluates multiple viewpoints and combines observations to reduce uncertainty, resolving ambiguities for approximately 92 percent of objects during testing.
AI-generated summary MariaDB on Wednesday added four agent-focused capabilities to Enterprise Platform 2027 to help developers retrieve data faster. The new features include AI Retrieval, AI Plugins, Agent Memory, and Semantic Cache for enterprise users.
AI-generated summary Amazon Web Services introduced Kiro bots on Thursday to oversee recurring development jobs and coordinate worker sessions. AWS states that the bots retain memory, maintain dedicated workspaces, and support models from Anthropic and OpenAI.
AI-generated summary LangChain discusses how agents navigate limited context windows by using retrieval techniques and agent harnesses. LangChain explains that tools like semantic search, filesystem indexing, and progressive disclosure help agents find necessary information.
AI-generated summary Perplexity is rolling out its new proactive agents called Teammates to Perplexity Max subscribers on the web. These agents operate within an isolated environment to execute computer tasks, skills, memory, and tools, with mobile app releases coming soon.
AI-generated summary MIT researchers have created a memory system called UQ-DAAAM to help robots identify and correct incorrect descriptions of objects stored in their memory. The system compares multiple descriptions of the same object seen from different viewpoints to quantify uncertainty and update captions in real time.
AI-generated summary Living Brain has launched from stealth with an external memory system that operates outside the model across different tools. The company states that the system achieves 92.2 percent on LoCoMo with a median retrieval time of 464 milliseconds.
AI-generated summary Baseten launches Baseten Sandboxes to provide stateful and secure environments for running agents. Baseten says Sandboxes boot from standby in milliseconds and are free to run until November 16.
AI-generated summary Salesforce made available an updated Agentforce Memory capability in September to help agents keep customer details current. Salesforce says the updated memory policy recalls more facts than the policy it replaced.
AI-generated summary Security researchers have released details about a technique known as memory poisoning, which allows attackers to inject malicious or misleading memory items into an artificial intelligence assistant's persistent memory layer. Forcepoint states that the aim of these attacks is to trick artificial intelligence agents into producing false information, creating a persistent risk that can repeatedly output flawed data across future sessions.
The top AI news in your email delivered each morning. Choose categories