Skip to content
AI News HubLIVE

This edition’s highlights

Agents

How to build evidence-grounded agent memory with OpenWiki and Graphs

  • OpenWiki compiles documents into a linked wiki with citations, serving as a persistent memory layer.
  • On connected questions, the wiki improved accuracy 5x over search alone and cut cost per correct answer.
LangChain BlogIn-site articleHow to build evidence-grounded agent memory with OpenWiki and Graphs

How Ropes & Gray is developing AI for transactional diligence

  • Ropes & Gray partners with OpenAI to develop an AI tool for transactional diligence.
  • The tool parses thousands of pages, groups documents by themes, and generates summaries and reports.
OpenAI NewsIn-site articleHow Ropes & Gray is developing AI for transactional diligence

Browser Use in Managed Deep Agents

  • Agents can perform web actions like clicking, scrolling, and form filling that APIs cannot cover.
  • Thread-scoped and user-scoped browser sessions are supported, with user scope persisting logins across conversations.
LangChain BlogIn-site articleBrowser Use in Managed Deep Agents

Voice Agents are here. Context turns them into real customer value.

  • Voice agents need customer context and business rules to deliver personalized service, not generic answers.
  • Databricks provides context, ontology, and governance as the brain, alongside fast decision models for real-time routing.
Databricks BlogIn-site articleVoice Agents are here. Context turns them into real customer value.

Building an Agentic AI Platform on Amazon Bedrock AgentCore

  • A federated operating model balances team autonomy with centralized guardrails.
  • Each workstream cell has isolated runtime, memory, and tool gateway for independent lifecycle management.
AWS Machine Learning BlogIn-site articleBuilding an Agentic AI Platform on Amazon Bedrock AgentCore
Models

How we detect frontier model misalignment

  • Monitoring covers training, evaluation, internal use, and external deployment for frontier models.
  • The multilayer system escalates from lightweight activation probes to model-based reviewers and investigative agents.
OpenAI NewsIn-site articleHow we detect frontier model misalignment

Distributed training with Ray Train DLCs and EFA on EKS

  • The Ray Train DLC already bundles CUDA, PyTorch, NCCL, EFA libraries, and the AWS OFI NCCL plugin, so no image build is needed
  • The sample trains Qwen2.5-1.5B with FSDP on eight L4 GPUs across two g6.12xlarge instances in a single Availability Zone
AWS Machine Learning BlogIn-site articleDistributed training with Ray Train DLCs and EFA on EKS

Quoting Ted Sanders

  • LLM performance is multidimensional with no fixed ordering, but model size explains much of the variation.
  • OpenAI models are ordered Luna < Terra < Sol < Astra, with bigger models being more reliable, costlier, and better at recalling esoteric facts.
Simon Willison's WeblogIn-site articleQuoting Ted Sanders
Policy

How Procore is powering construction analytics with OpenSharing

  • Procore Analytics 2.0 uses Databricks medallion architecture and OpenSharing to distribute hundreds of curated tables
  • Refreshes increased from daily to hourly, with support for hundreds of thousands of distinct data shares
Databricks BlogIn-site articleHow Procore is powering construction analytics with OpenSharing

You have reached the end of this edition.

Your reading list
Other updates (5)
Agents

Quoting Ben Hylak

Ben Hylak ironically argues that every product should become a headless MCP feeding an agent—except his own, whose customers love its UI. Collected by Simon Willison on October 29, 2026, the quote captures the tension between MCP-driven agents and interface design.

Simon Willison's WeblogIn-site articleQuoting Ben Hylak

Credit Saison achieves 954% ROI in a ChatGPT Enterprise pilot

Credit Saison's ChatGPT Enterprise pilot delivered a 954% return on investment and saved participants 14 hours per month, leading the company to approve a companywide rollout just one month after the pilot ended.

OpenAI NewsIn-site articleCredit Saison achieves 954% ROI in a ChatGPT Enterprise pilot

How to Build Proactive, Always-On Agents with the Schedules SDK in Managed Deep Agents

Most proactive agents run on a simple cron. This article explains what scheduled runs unlock, from memory to smart triage, and how to add them to Managed Deep Agents. More than 70% of Managed Deep Agent runs on LangSmith come from scheduled runs rather than human messages.

LangChain BlogIn-site articleHow to Build Proactive, Always-On Agents with the Schedules SDK in Managed Deep Agents
Policy

How Procore is powering construction analytics with OpenSharing

Procore replaced its API-driven, once-daily refresh analytics product with OpenSharing on Databricks, doubling data availability, lifting refresh frequency to hourly, and scaling to hundreds of curated tables for construction customers.

Databricks BlogIn-site articleHow Procore is powering construction analytics with OpenSharing