Skip to content
AI News HubLIVE

This edition’s highlights

Agents

Gondola doubles revenue as its team builds more with Codex

  • Codex increased overall engineering output by 25% and lead engineer's merged changes by 50% with no headcount growth.
  • Product managers and business staff use Codex Cloud to prototype ideas and even make their first code contributions.
OpenAI NewsIn-site articleGondola doubles revenue as its team builds more with Codex

Holiday Extras is rethinking what it can build with Codex

  • Codex enabled a two-week rebuild of Holiday Extras' main app that would previously have taken over six months
  • Open-to-book conversion rose 32.5% and the app error rate fell from 19% to 9% after the rebuild
OpenAI NewsIn-site articleHoliday Extras is rethinking what it can build with Codex
Models

Gemini Decisions API: Fast, calibrated multimodal decisions

  • Gemini Decision 1 scores candidate options in one forward pass instead of token-by-token decoding, achieving 150-250 ms end-to-end latency.
  • Each request can combine up to 1,000 independent questions over one shared multimodal input of up to 256K tokens, using yes/no, multiple-choice, and ordered rubric question types.
Google AI BlogIn-site articleGemini Decisions API: Fast, calibrated multimodal decisions

ParseBench-2: Severity-aware document parsing benchmark

  • Introduces severity-tuned scoring, using 330+ error types to separate critical factual errors from harmless formatting differences
  • Expands the dataset to 4.3k cases, adds forms as a sixth dimension, and evaluates 36,958 chart values
LlamaIndex BlogIn-site articleParseBench-2: Severity-aware document parsing benchmark

Baseten joins the NVIDIA Nemotron Coalition to advance custom AI in production

  • Baseten joins the NVIDIA Nemotron Coalition to expand collaboration on open models and enterprise AI adoption
  • Baseten builds RL environments for Nemotron targeting complex instruction following, long-horizon reasoning, and multimodal tasks
Baseten BlogIn-site articleBaseten joins the NVIDIA Nemotron Coalition to advance custom AI in production

llm 0.37.2 Released: Fixes Two Security Issues That Could Leak OpenAI API Keys

  • llm 0.37.2 addresses security advisories GHSA-v9h7-jj4g-f862 and GHSA-r639-rr7r-x72w.
  • The vulnerabilities affect the extra-openai-models.yaml mechanism for connecting to third-party OpenAI-compatible APIs.
Simon Willison's WeblogIn-site articlellm 0.37.2 Released: Fixes Two Security Issues That Could Leak OpenAI API Keys
Research

Using wearables to enable cardiovascular-kidney-metabolic (CKM) risk profiling for personalized coaching

  • The coachable health age (CHA) model uses only modifiable daily behaviors and physiology (resting heart rate, steps, sleep, active zone minutes, etc.) and was validated across three independent cohorts, including All of Us, to predict future disease risk.
  • SensorWM, a world model pre-trained on 2.28 million person-weeks from about 100,000 Fitbit users, simulates how 20 physiological markers evolve over 12 weeks and is paired with an optimizer and language model to generate safe, feasible coaching plans.
Google Research BlogIn-site articleUsing wearables to enable cardiovascular-kidney-metabolic (CKM) risk profiling for personalized coaching
Chips

A New Dawn in the Cloud: ‘Warhammer 40,000: Dawn of War IV’ Is Coming to GeForce NOW

  • Warhammer 40,000: Dawn of War IV launches on Steam December 3 and comes to GeForce NOW, returning to large-scale battles and base building.
  • War Thunder and Where Winds Meet will add DLSS 5 with 3D-Guided Neural Rendering, with the latest model doubling performance over the previous version.
NVIDIA BlogIn-site articleA New Dawn in the Cloud: ‘Warhammer 40,000: Dawn of War IV’ Is Coming to GeForce NOW

You have reached the end of this edition.

Your reading list
Other updates (31)
Agents

Build composable, interoperable multi-agent systems with A2A on AWS

AWS published a reference architecture showing how to build composable, interoperable multi-agent systems across organizational boundaries using the Agent-to-Agent (A2A) protocol. A LangChain orchestrator on Amazon ECS communicates with a Strands agent on Amazon Bedrock AgentCore through bidirectional, OAuth-authenticated A2A calls, with neither side depending on the other's framework or SDK.

AWS Machine Learning BlogIn-site articleBuild composable, interoperable multi-agent systems with A2A on AWS

From prototype to production: Engineering agents with Agent DLC

AI agents moving from demo to production face non-determinism, prompts-as-source-code, and dependency drift. Agent DLC proposes one lifecycle, two loops, and a single explicit acceptance standard backed by evaluation evidence. A measured example shows aggregate metrics can hide retrieval coverage gaps, and prompt changes alone cannot fix missing knowledge.

AWS Machine Learning BlogIn-site articleFrom prototype to production: Engineering agents with Agent DLC

Genie in Slack is now Generally Available: Governed AI cowork, right where work happens

Databricks announces the general availability of the Genie app for Slack, enabling users to ask questions and get governed answers directly in Slack channels or DMs, with the ability to take actions via MCP connectors and schedule analyses, all while respecting Unity Catalog permissions.

Databricks BlogIn-site articleGenie in Slack is now Generally Available: Governed AI cowork, right where work happens

Build autonomous AI agents with Amazon Bedrock and Spring AI

Learn how Java developers can build production-ready autonomous AI agents using Amazon Bedrock and Spring AI, covering tool calling, conversation memory, resilience patterns, and multi-agent orchestration, then deploy to AWS with AgentCore.

AWS Machine Learning BlogIn-site articleBuild autonomous AI agents with Amazon Bedrock and Spring AI

Amazon Bedrock AgentCore adds a standalone Web Search API

AWS introduced a Web Search API on the Bedrock AgentCore data plane, letting developers call InvokeWebSearch directly with standard IAM credentials instead of provisioning an AgentCore Gateway. It returns ranked web results with extracts and metadata, supports domain and date filters, and keeps queries inside AWS while offering PrivateLink and CloudTrail controls.

AWS Machine Learning BlogIn-site articleAmazon Bedrock AgentCore adds a standalone Web Search API

ICYMI: What landed for AI builders in October 2026

AWS recaps October 2026 updates across Amazon Bedrock, AgentCore, and Strands, covering new model choices, agent reliability and availability improvements, and clearer cost reporting.

AWS Machine Learning BlogIn-site articleICYMI: What landed for AI builders in October 2026

AWS ships official aws-emulator, built by an agent swarm with 29,684 commits in three weeks

AWS has released aws-emulator, an official local emulator that lets developers test against clones of many AWS services without provisioning a paid API key. The Rust-based project was developed by a large swarm of AI agents, producing 29,684 commits in its first three weeks. It currently requires Docker on macOS and Windows.

Simon Willison's WeblogIn-site articleAWS ships official aws-emulator, built by an agent swarm with 29,684 commits in three weeks

The attendee’s guide to the AWS re:Invent 2026 AI/ML track

AWS re:Invent 2026, running November 30 to December 4, features more than 140 AI/ML track sessions. This guide organizes them by the AWS AI stack—AI chips, models and inference, agents, and cross-cutting security, governance, and cost optimization—to help attendees plan a schedule that fits their role and goals.

AWS Machine Learning BlogIn-site articleThe attendee’s guide to the AWS re:Invent 2026 AI/ML track

Automating AWS Agent Registry updates with push-based architecture

This post shows how to keep the AWS Agent Registry current automatically. Using AWS CloudTrail, Amazon EventBridge, AWS Lambda, and AgentCore Identity, an event-driven pipeline detects MCP server changes and pushes updated tool metadata to the registry with no stored secrets, across single- and multi-account environments.

AWS Machine Learning BlogIn-site articleAutomating AWS Agent Registry updates with push-based architecture

How DXC Technology built a multi-model AI substrate on Amazon Bedrock

DXC Technology built DXC OASIS, a multi-model AI substrate on Amazon Bedrock, separating a governed model layer from a trusted harness for mission-critical IT operations, achieving 95% faster resolution times.

AWS Machine Learning BlogIn-site articleHow DXC Technology built a multi-model AI substrate on Amazon Bedrock

Accelerating generative AI delivery with reusable assets

AWS Generative AI Innovation Center offers 23 production-ready open source reusable assets to help teams avoid rebuilding components like RAG pipelines, document processors, and agent scaffolding. Companies such as DoorDash, Lyft, and Ricoh achieved significant results, including 50x testing capacity and 90% reduction in engineering hours.

AWS Machine Learning BlogIn-site articleAccelerating generative AI delivery with reusable assets

How our GTM Agent builds consistent, high-quality sales artifacts

LangChain's GTM Agent, built on Deep Agents, automates the creation of customized sales documents. Quality comes mainly from business context rather than prompt tuning. Over 12 weeks, it generated 1,540 artifacts, saving thousands of hours; deals with a sent artifact advanced 47% of the time vs. 16% for matched deals without one.

LangChain BlogIn-site articleHow our GTM Agent builds consistent, high-quality sales artifacts

Air gapped, not data dark: how Bright Machines gives customers live factory dashboards without opening the plant

Bright Machines captures PostgreSQL changes at the factory edge with Debezium, writes them as Apache Iceberg, sends them over a single auditable outbound path, and reads them in place in Databricks through Unity Catalog federation, giving customers live governed dashboards without any inbound access to air-gapped plants.

Databricks BlogIn-site articleAir gapped, not data dark: how Bright Machines gives customers live factory dashboards without opening the plant

Accelerating product development with AI-driven lifecycle management (AI-PDLC) in AWS Quick

AWS blog introduces an AI-driven product development lifecycle (AI-PDLC) framework built on AWS Quick, featuring six specialized agents that cover the Discover, Decide, Design, and Prototype phases with human approval gates, producing a specification ready for coding agents like Kiro.

AWS Machine Learning BlogIn-site articleAccelerating product development with AI-driven lifecycle management (AI-PDLC) in AWS Quick

Building the Agent Cloud: Introducing Baseten Sandboxes

Baseten launches Baseten Sandboxes, a microVM-based execution environment for AI agents with instant pause/resume (about 25 ms), hardware-level isolation, and elastic scaling. It is the first piece of a purpose-built agent cloud, with Agent Drive storage and Agent Proxy networking coming soon.

Baseten BlogIn-site articleBuilding the Agent Cloud: Introducing Baseten Sandboxes
Models

Fine-grained access control for Amazon Bedrock Managed Knowledge Base tools

This post shows how to expose an Amazon Bedrock Managed Knowledge Base to AI agents as MCP tools through Amazon Bedrock AgentCore Gateway while keeping per-user access control. Trusted code, not the LLM, sets the user identity: a request interceptor injects the verified identity, Policy in AgentCore validates it independently, and a resource-based policy restricts direct access to the knowledge base.

AWS Machine Learning BlogIn-site articleFine-grained access control for Amazon Bedrock Managed Knowledge Base tools

How Block powers collaborative AI agents with Buzz and Unity Gateway

Block developed Buzz, an open source collaboration space where employees and AI agents work together. Using Databricks Unity Gateway, Buzz provides unified access to frontier and open-weight models, enabling multi-agent workflows and model efficiency experiments.

Databricks BlogIn-site articleHow Block powers collaborative AI agents with Buzz and Unity Gateway

Theoretical Limits of Language Model Alignment

This paper characterizes the information-theoretic limits of KL-regularized alignment, deriving a closed-form expression for optimal reward improvement governed by Jeffreys divergence rather than √KL, and shows best-of-N approaches the limit while PPO and GRPO are suboptimal.

Apple Machine Learning ResearchIn-site articleTheoretical Limits of Language Model Alignment

Gemini Decisions API

Google launches Gemini Decision 1, joining Jev and OpenAI in the decision model category. It supports yes/no, choice, and rubric-scoring questions across text, images, audio, video, and PDFs, priced like GPT-6 Luna at 10 cents per million input tokens with free output.

Simon Willison's WeblogIn-site articleGemini Decisions API

ElevenLabs Summit NYC: The Interaction Layer for AI

At its New York Summit, ElevenLabs unveiled domain-specific models, fine-tuning for open-source and voice models, the Interaction Engine for orchestrating conversations, new ElevenAgents features like Memory and Routines, the ElevenAI enterprise app, and an expanded partner program and marketplace to build the interaction layer for enterprise conversational AI.

ElevenLabs BlogIn-site articleElevenLabs Summit NYC: The Interaction Layer for AI

The platform for inference, training, and execution | Baseten

Baseten Applied Research and Genspark post-trained Qwen3.5-397B into a specialized deep research model that scores 0.759 on Genspark's grader versus 0.612 for the base model, beating Opus 5.5, Sonnet 5.5, GPT-5.6 Sol, Kimi K3 and DeepSeek V4 Pro. It crawls 5x more sources per query, runs with NVFP4 weights on just 2x B200 GPUs per replica, and now powers Genspark's Deep Research agent in production.

Baseten BlogIn-site articleThe platform for inference, training, and execution | Baseten

Baseten launches Serverless Loops: frontier RL training for everyone

Baseten announced the general availability of Serverless Loops, a training SDK for supervised fine-tuning and reinforcement learning on frontier open models with a one-click path to production inference.

Baseten BlogIn-site articleBaseten launches Serverless Loops: frontier RL training for everyone
Research

Supporting research on AI and children’s safety

OpenAI signed the Charter of Engagement for Safer AI for Children at the Paris Peace Forum and committed $2 million to the research network under the Coalition for Children’s Rights and Protection in the Age of AI, focused on shared tools for measuring risks and opportunities for young people.

OpenAI NewsIn-site articleSupporting research on AI and children’s safety
Policy

Gilbert + Tobin rethinks the business of law with AI

Australian law firm Gilbert + Tobin is scaling AI through ChatGPT Enterprise, ChatGPT Work and Codex, moving from individual experimentation to governed, multistep agentic workflows. CEO Sam Nickless says the value lies in freeing people for more human work, not cutting costs or headcount.

OpenAI NewsIn-site articleGilbert + Tobin rethinks the business of law with AI

OpenAI Codex on Amazon Bedrock

A practical guide for enterprises rolling out OpenAI Codex to developers on AWS, with Amazon Bedrock as the model backend, covering the access, governance, and cost decisions that platform, security, and finance teams need to align on.

AWS Machine Learning BlogIn-site articleOpenAI Codex on Amazon Bedrock

The building blocks of the open frontier

Baseten launches Grid, Sandboxes, and Loops to unify inference, training, routing, governance, and sandboxing on one platform for building and running AI systems.

Baseten BlogIn-site articleThe building blocks of the open frontier
Chips

Detecting and recovering from silent data corruption on Amazon SageMaker HyperPod

Silent data corruption (SDC) is a GPU fault that returns wrong but plausible numbers with no error, quietly poisoning distributed training. This post adds three checks to a PyTorch DDP loop to detect it, localize the faulty node on Amazon SageMaker HyperPod, and hand the node to automatic recovery for replacement.

AWS Machine Learning BlogIn-site articleDetecting and recovering from silent data corruption on Amazon SageMaker HyperPod
Tools

Browse tools.simonwillison.net by tag

Simon Willison added tag-based browsing to his collection of 244 vibe-coded tools. Using leftover Fable 5.2 credits, he employed tree-sitter to analyze each tool's JavaScript APIs, libraries, and web API hosts, generating tag clouds by category on the homepage.

Simon Willison's WeblogIn-site articleBrowse tools.simonwillison.net by tag
Robotics

The new Genie: autonomous AI for complex enterprise work

Databricks introduces the new Genie, an autonomous AI coworker that handles long-running enterprise tasks within user-set guardrails. It uses Genie Ontology for business context, Genie Search for hybrid search across the lakehouse and business systems, and Unity Gateway for admin control over tool access, auditing, and spend. Genie works through Genie One, email, Slack, and Teams, supports parallel jobs, proactive task suggestions, artifacts, and MCP connectors. Genie Agents bring specialized expertise to teams, Genie Ontology is built by agents and governed by people, and Genie Search enforces access controls. A benchmark with Turing showed 91% accuracy and less than half the tokens of leading frontier agents.

Databricks BlogIn-site articleThe new Genie: autonomous AI for complex enterprise work
Startups

Working at an early stage company as an early stage engineer

A new grad engineer shares their journey from college senior to early engineer at Baseten, covering job hunting, interviews, and deciding to join a startup.

Baseten BlogIn-site articleWorking at an early stage company as an early stage engineer