The latest AI news, in focus.

Microsoft open-sources Agensh: a decentralized framework that supports 1024 autonomous, collaboratively operating AI coding agents.

On October 25, 2026, Microsoft Research open-sourced Agensh, a multi-agent runtime framework that operates without a central coordinator, enabling 1 to 1,024 LLM-powered coding agents to self-organize and collaborate on ProgramBench. The system achieves decentralized coordination through shared dashboards inspired by Gitea, Mattermost, and DeLM, with all agents following the same prompt-driven five-step loop; its scalability has been validated in a Linux + Docker environment.

Why it mattersThis achievement represents a significant evolution in multi-agent system architectures—abandoning the traditional “master‑agent” paradigm and enabling large‑scale self‑organization through prompt engineering and lightweight service composition. Empirical results on ProgramBench (e.g., 128 agents achieving a 49% improvement in task load balancing across five tasks) establish a new benchmark for distributed AI engineering. Released under the MIT License and compatible with mainstream single‑agent tools such as Copilot CLI, it offers valuable guidance for practical development.

Taking on NVIDIA head-on to challenge Apple—Microsoft’s high-end AI PC gamble

Microsoft is courting local AI developers and high-end creators with the Surface Laptop Ultra, which features NVIDIA RTX Spark. The hardware brings large-capacity unified memory, CUDA, and RTX graphics capabilities to a portable device, but whether it can sustain competition against Macs will still depend on app optimization, agent-based experiences, and real-world workflows.

Why it mattersUltimately, this competition comes down to whether on-device AI can genuinely boost creativity and development efficiency, and whether the associated costs—hardware, software, and migration—are justified. Hardware opens up new possibilities, but the ecosystem and real-world user experience determine how many users those possibilities can reach.

Anthropic open-sources the OS-level sandbox sandbox-runtime: providing container-free file and network isolation for AI agents.

On October 24, 2026, Anthropic open-sourced sandbox-runtime (srt), an OS-level sandboxing tool that operates without containers, designed to restrict file read/write and network access permissions for AI agents, MCP servers, and CLI tools. It supports macOS (Seatbelt), Linux (bubblewrap), and Windows (alpha), with npm package version 0.0.81 under the Apache-2.0 license.

Why it mattersThis tool represents a significant advancement in AI security infrastructure—achieving fine-grained, full-process-tree coverage through the operating system’s native isolation mechanisms, directly addressing the risks of prompt injection and privilege escalation, and has already been deployed for internal validation within Claude Code.

Letta platform releases a tutorial on building stateful engineering collaborators: supporting core memory, shared state, and portable workflows.

MarkTechPost published a tutorial on October 24, 2026, detailing how to build an agentic copilot with persistent state using the Letta framework. The guide covers capabilities such as verifiable core memory, multi-agent shared memory, semantic retrieval of archived knowledge, custom Python tool integration, streaming responses, and .agent format (.af) export. The content serves as a technical practice guide and does not provide performance metrics, security audits, or production‑environment validation.

Research reveals the “norm–oracle gap” problem in benchmarking coding agents.

A paper published on arXiv on October 24, 2026, points out that current mainstream benchmarks for coding AIs—such as SWE-Milestone—suffer from a “specification–oracle gap”: the natural-language descriptions of tasks fail to capture crucial implementation details required for testing (e.g., constant names, module paths, and precise error messages), resulting in evaluations that measure the ability to reproduce upstream code rather than genuine understanding and reasoning.

Why it mattersThis study is the first to formally define this type of bias, develop a static analysis tool covering five languages (with 89.2% precision and 92.9% recall), and empirically demonstrate in SWE-Milestone that 25 milestone tasks contain non-inferable implicit dependencies, significantly increasing the failure rate (80.6% vs. 48.4%). These findings reveal that current leaderboards may systematically favor “name‑copying” agents. The results have been made publicly available, including the open-source auditing tool and corrective resources, offering both methodological and practical value.

Amazon SageMaker AI has launched a Multi-Step Reinforcement Learning (MTRL) framework, supporting end-to-end training of agents to accomplish multi-step tasks.

AWS has launched the Multi-Turn Reinforcement Learning (MTRL) feature, enabling users to train large‑model agents on SageMaker AI using full task trajectories and final outcome feedback, without requiring step‑level annotations. A end-to-end demonstration is provided using Wordle as an example, covering AgentCore deployment, data preparation, GRPO training, reward design, and cost analysis.

Bayer Crop Science has built the multi-agent decision-making engine ADE based on Amazon Bedrock AgentCore.

Bayer Crop Science has built the Agentic Decision Engine (ADE) on Amazon Bedrock AgentCore—a natural-language multi-agent platform for breeding experts, analysts, and researchers—as an enhancement layer rather than a replacement for its existing Connected Decision Engine (CDE). The system is now live in production, serving over 100 R&D users and enabling approximately 80% of issues to be resolved independently through 22 domain-specific agents and about 20 API tools.

Black October: A Comprehensive Overview of the AI-Driven Cybersecurity Crisis—From Its Peak to Defensive Response

Around October 21–22, 2026, over a span of approximately 48 hours, more than a dozen high‑severity AI‑related vulnerabilities and intrusion incidents were reported worldwide, including PHP2Shell, GitLab RCE (CVSS 10), NGINX/HTTP/3 DoS, LiteLLM password theft, eMudhra CA key leakage, and Quest Diagnostics medical data exposure. State actors such as Storm‑0940 leveraged open‑source intelligent penetration‑testing frameworks like ARTEX to launch large‑scale attacks. Concurrently, AI security defense systems were rapidly being developed, encompassing formal verification, open‑source patching agents, standardization of agent capabilities, and hardware‑level optimizations.

Canva has launched its first in-house image-generation model, Canva Craft, which is now available to all users.

Canva has officially launched its first in-house image-generation model, Canva Craft, which employs a two-stage generative architecture, supports 2K resolution and output in 30 aspect ratios, and emphasizes text readability and layer editability. The model was trained on millions of human-designed templates within the platform and has already been integrated into Dream Lab and the API; going forward, it will be accessible via MCP to third-party chatbots such as ChatGPT and Gemini.

The Cornerstone of Scalable AI: Building Responsible Systems in Regulatory Environments

MIT Technology Review’s EmTech Future conference delves into the five pillars for building scalable AI systems in a heavily regulated environment—data, engineering, platforms, business models, and governance—and underscores how trustworthy, responsible deployment can transform emerging capabilities such as agentic AI into sustainable investment value.

Donating Agent Skills to the Agentic AI Foundation

Donating Agent Skills to the Agentic AI Foundation News today from Anthropic: We’re committed to keeping Agent Skills open, community-driven, and vendor-neutral. Today, we are donating the standard to the Agentic AI Foundation (AAIF) under the Linux Foundation, which we co-founded with Block and OpenAI last year, and to which we donated the Model Context Protocol (MCP). I'm pleased to note I called this one last December, when Anthropic first declared skills to be an "open standard": I wouldn't be surprised to see this end up in the AAIF, recently the new home of the MCP specification. Fun side note: since my blog content is available to Claude via the datasette-mcp plugin I found that post by prompting Claude to "Use simonwillison.net MCP to find if I ever talked about how I want skills t

Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal CEO

Elastic GPUs, sandbox growth, and the economics of the agent cloud

Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal CEO

Elastic GPUs, sandbox growth, and the economics of the agent cloud

Spotify bets on the AI assistant era: becoming the music service of choice for AI assistants

Spotify co-CEO Gustav Söderström stated during the Q3 2026 earnings call that the rise of AI agents will not diminish the value of streaming platforms; on the contrary, it will reinforce Spotify’s unique advantage—its vast trove of personalized listening data and deep agent integration capabilities, making it the preferred music service interface for third-party AI assistants.

A Multi-Layered Approach to AI Agent Hallucination Detection: A Practical Framework for Production Environments

This paper systematically categorizes four typical types of hallucinations exhibited by AI agents in real-world workflows—factual, contextual, fictional, and self-contradictory—and proposes that detection should rely on a multi-dimensional approach, incorporating credibility checks, source verification, self-consistency assessment, semantic entropy analysis, and human–machine collaboration thresholds, rather than on a single scoring metric. It underscores that such detection must be embedded within the workflow’s closed loop and deeply integrated with observability, error handling, and human review pathways.

Microsoft and Hugging Face have open-sourced the ThinkingBox framework: using the database’s final state as a deterministic reward signal to advance post-training reinforcement learning for agents.

Microsoft and Hugging Face have jointly launched the open-source framework ThinkingBox, which generates binary 0/1 rewards by comparing the LLM agent’s final database state after execution with the expected state, for reinforcement learning-based fine-tuning. The accompanying benchmark, ThinkingBox-Bench, includes 507 real-world backend tasks, none of which were used during training. After 50 steps of GRPO fine-tuning, Qwen3.8-27B achieved a 9.08-percentage-point improvement in pass@1 on the held-out test set, with Observed 20/20 nearly tripling; however, pass@20 declined in some domains, reflecting a trade-off between breadth and stability.

Deep Insight: How AWS’s Open-Source Multi-Agent System Addresses Five Major Production-Grade Challenges

AWS Solutions Architecture Team has launched Deep Insight, an open-source multi-agent system designed for automated CSV data analysis and natural-language report generation. It systematically addresses five common challenges in production environments—agent orchestration, role-based model selection, stable operation, secure code execution, and observability—and has been validated in real-world scenarios such as LG Electronics and the re:Invent 2025 TechRecon.

AWS releases the Adaptive Agent Framework (AAIF): a hierarchical AI task scheduling solution that enables edge-cloud collaboration based on Strands Agents.

AWS has introduced the Adaptive Agentic Intelligence Framework (AAIF), which leverages Strands Agents to build a hierarchical agent architecture. It dynamically routes tasks via local small language models (SLMs): routine IoT telemetry and visual analytics are executed at the edge, while complex reasoning is offloaded to Amazon Bedrock’s large cloud-based models. The framework supports hybrid deployment, keeps costs under control, and incorporates responsible AI governance recommendations; however, it currently exists as a proof-of-concept implementation and does not yet provide third-party validation or commercial deployment data.

Meta confirms that the AI app Muse has surpassed 10 million users, with shopping emerging as its core use case.

During its Q3 2026 earnings call, Meta confirmed that the number of users of its AI agent app, Muse, has surpassed 10 million; approximately 25% of new users engage with it daily, with shopping being the most frequent activity—over a quarter of active shopping users spend more than $200 per transaction on average. Meanwhile, Plaid reported that its payment connector’s API calls on Muse have reached an all-time high.

Walmart launches an AI-powered shopping gateway, supporting autonomous shopping workflows and enhancing user control.

Walmart has officially launched Agent Gateway (an AI agent gateway) along with its accompanying Agent Profiles feature, enabling authorized AI agents—such as ChatGPT, Gemini, Meta Muse, and others—to perform searches and place orders on behalf of users across Walmart and Sam’s Club platforms. User control over agent activities is ensured through mechanisms like independent account configuration, transaction approvals, and spending limits. This move stands in stark contrast to Amazon’s recent ban on AI‑powered shopping agents.