Search “Agents”

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

AWS Solutions Architecture team builds multi-agent system Deep Insight

The AWS Solutions Architecture team built and open-sourced the multi-agent system Deep Insight, which can transform CSV data and natural language prompts into cited .docx analysis reports. The system consists of eight agents—Coordinator, Planner, Supervisor, Coder, Validator, Reporter, Tracker, and others—collaborating in a layered manner, controlling execution flow based on the Graph and Agents-as-Tools patterns of the Strands Agents SDK, and selecting Claude Haiku, Opus, and Sonnet respectively by role. Production deployment relies on the Amazon Bedrock AgentCore runtime, with LLM-generated code executed in isolation within per-session ephemeral containers on AWS Fargate, accompanied by a Web UI and Ops dashboard.

Google DeepMind releases the Gemini 4 Carbon model

On October 28, 2026, Google DeepMind released Gemini 4 Carbon, calling it its strongest model, ranking first with 59 points on the Artificial Analysis Intelligence Index v5.0, and scoring 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench. The model is aimed at long-horizon engineering tasks and was first opened to some cybersecurity defenders, then expanded through the Fairwind Program, and afterward made available to developers, enterprises, and consumers. On October 29, Arena.ai said Gemini 4 Carbon ranked 2nd in Agent Arena with a net improvement of 16.2%, and was on the Pareto frontier for cost, time, and output tokens per task; the same day, Yuchen Jin cited Artificial Analysis data saying it was smarter than Opus 5.5 and cheaper per task, but noted that the evaluation used the Antigravity CLI. On October 30, the Artificial Analysis Intelligence Index showed it in first place with 59 points, 1 point higher than Claude Opus 5.5. On October 31, Perplexity CEO Aravind Srinivas shared early evaluations

DeepLearning.AI Launches Context Engineering Course

On October 29, 2026, DeepLearning.AI launched a Context Engineering course, taught by Chroma CEO Jeff Huber and CMPND CEO Drew Breunig, with context engineering as the throughline for keeping agents reliable in complex multi-step tasks and saving tokens. The course builds a harness from scratch in Python and uses traces and evals to quantify each change, covering three capabilities: Agentic Search, Context Management, and Agentic Memory. It is divided into four modules, and the labs are mostly code exercises of about one hour. The course assumes intermediate Python, familiarity with LLM applications and tool calling, and basic retrieval. There are currently no further developments.

Addy discusses how to use Claude Code in the era of Claude Opus 5.5

Addy Osmani, a member of the Anthropic Claude Code team, published a post on October 30, 2026, discussing how to use Claude Code in the era of Claude Opus 5.5. He proposed that starting with the Claude 5 series, users should reduce piling up prompts and hand-holding task breakdowns, and instead use a single prompt to clearly state the goal and acceptance criteria, letting the model run to completion on its own and then stop; reasoning effort can start from /effort medium, and when the goal is unclear, first have Claude interview you to produce a spec, then open a new session to implement it. He mentioned that Claude Code once removed more than 80% of its system prompts for models such as Claude Opus 5 and Claude Fable 5, with no measurable decline in coding benchmarks, and recommended using /doctor prompt-audit to find prompts written for older models. He also said that the Claude Code interactive terminal recently defaulted to auto mode, but this is not a safety guarantee; subagents should be split manually as little as possible, and a single agent plus skills often yields better quality.

GitHub announces that Copilot Fusion is officially available

On October 29, 2026, GitHub announced that Copilot Fusion is officially available. According to a report by Rohan Paul, the product can route coding tasks between Claude and GPT models; on Artificial Analysis's Coding Agent Index, its Max setting scored 60, compared with 62 for Claude Code, at a cost per task of $5.59 versus $11.42, with Fusion consuming 7.2M tokens per task, roughly half of Claude Code's usage. On the same day, Microsoft released multiple updates at GitHub Universe, stating that Copilot Fusion achieves frontier-level quality, is 40% faster, and reduces costs by up to 77%, and announced that a unified Copilot harness is entering Foundry as a managed service. There are currently no subsequent developments.

blakeandersonw open-source ai-app-playbook

On October 28, 2026, Blake Anderson open-sourced "ai-app-playbook," a set of operating procedures for coding agents to execute, covering product selection, building, publishing, and first-month profitability. The playbook has two layers: a human-facing playbook with chapters 00–13, and an Agent-facing AGENTS.md (Phase 0–14, with each step marked [AGENT] or [HUMAN]); before starting, one must pass the Gate A scoring. The tech stack is fixed as Cloudflare Workers, D1, Better Auth, OpenRouter routing to DeepSeek, Superwall plus Stripe, etc., and explicitly lists items not to use, such as Vercel, Supabase, and Next.js. Currently there is only this open-source release information, with no subsequent developments yet.

OpenAI launches GPT-6.1 Terra on ChatGPT Work and Codex

On October 27, 2026, OpenAI released and launched GPT-6.1 Terra, touted for its speed and efficiency, visual reasoning, and precise computer operation, with clearer and more natural writing as well, rolling it out that day to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. On the same day, Arena announced that the model had entered Agent Arena and Code Arena; OpenRouter launched the model, saying it matches GPT-6 Astra on Agents' Last Exam at one-tenth the price per token; OpenClaw also announced that the model is now selectable. OpenAI co-founder Greg Brockman said Terra's computer use has crossed the threshold of a qualitative change and that using it is faster than operating a computer on his own. fal, meanwhile, used the fal Codex plugin to test turning Blender 3D content into Seedance 2.5 cinematic footage. On October 28, Arena announced that Terra ranked 14th in Code Arena: WebDev with a score of 1671, at a median cost of $0.40 per task, 28% lower than GPT-5.6 Terra. OpenAI's system card confirms that Terra is overall safer than older models, but its flag rates on items such as concealing uncertainty, circumventing restrictions, data exfiltration, and destructive operations

TinyFish Releases Technical Guide for Production-Grade Agentic Search Architecture

On October 23, 2026, TinyFish released two technical guides: one defines a production-grade agentic search architecture, and the other summarizes lessons learned from building a deep research system. According to a MarkTechPost report, the architecture is built on four web primitives—TinySearch, TinyFetch, TinyBrowser, and TinyAgent—and runs a seven-stage control loop around a foundation model, covering task definition, retrieval planning, tool routing, evidence evaluation, and failure retry. In OpenBenchmarks tests with a fixed gpt-5.6-sol model and 100 tasks, the search-plus-fetch configuration achieved a 79.0% completion rate, with a median task cost of $0.050 and a runtime of 24 seconds; the search-only configuration achieved a 59.3% completion rate. Currently, this event has only the single source report mentioned above, and there have been no subsequent developments.

Simon Willison introduces the vibe-coded RTS game Battle Drones

Simon Willison introduced Battle Drones, a real-time strategy game he developed entirely through vibe coding with Claude Code. According to the information he published, the project went through 12 major versions and 321 commits and runs in mobile and desktop browsers. The game includes 9 building types, 7 ground vehicles, 6 air vehicles, and 3 infantry units, and supports fog of war, random map generation, AI versus AI, and an Agent battle mode in which two LLMs fight each other. He also implemented a multiplayer method in which two devices synchronize game state via microphone audio, and because the audio caused headaches, he added online multiplayer based on a WebSockets relay server. There are currently no subsequent independent developments in this story.

Stanford releases public video and lecture notes for CME 295 Lecture 4

Stanford course CME 295 Transformers & Large Language Models has been updated to Lecture 4, with both the video and lecture notes now publicly available. The topic of this lecture is Reinforcement learning with LLMs, covering everything from reinforcement learning and reward design to PPO, GRPO, RLVR, and DAPO, explaining each of the notations one by one. The course is taught by Afshine and Shervine Amidi, with 9 lectures in total, covering the Transformer architecture, attention variants, RoPE, LoRA fine-tuning, RLHF/GRPO training, Flash Attention inference optimization, MCP and AI Agents, and diffusion language models. Currently, Lecture 4 is public, and no evidence of updates for subsequent lectures has been seen yet.

Shao Meng recommends MIT course MAS.S62 Lectures 4-5: Harness Evolution

On October 20, 2026, Shao Meng recommended Lectures 4-5 of the MIT course MAS.S62 "How to AI an AI," themed "Harness evolution." According to his introduction, these two lectures respectively survey the evolution of harness components, memory, tools, and skills, as well as the evolution of planning, task decomposition, and multi-agent systems. The memory section uses a three-axis taxonomy to connect MemGPT, mem0, MEM1, and Reflexion/ExpeL, and uses CLM to unify context management as file operations; the tools section moves from Toolformer, ToolLLM, and CodeAct to Meta-Harness, which represents the entire harness as a file system, with feedback information roughly 400 times that of previous methods. The multi-agent section presents five empirical findings, including that MAS wins 81% on decomposable tasks and loses 70% on sequentially dependent tasks, and that CORAL achieves new SOTA on 8 tasks, with multi-agent evolution improving 4%–20% over single-agent. Currently, this recommendation has only one source published by Shao Meng himself, with no subsequent independent reports.

MIT OpenCourseWare 6.S950 updates Lecture 5 Harness handout

MIT OpenCourseWare 6.S950 Agency with AI has updated the Lecture 5 The Harness handout. The handout uses a teaching Agent Baby Anigans—about 300 lines of Python, containing only two tools, search and read_lines, and running locally on Ollama—as an example to break down how the Harness works, and summarizes the Agent as LLM plus Harness. The handout also discusses the transparency trade-offs in observability, namely the problems brought about after vendors encapsulate context summaries and reasoning processes into encrypted APIs, and traces practices such as context management, mandatory interception, and least privilege back to system principles from 1965 to 1975. The handout also points out that there is no universal optimal solution: among 5 models, 4 have different best Harnesses on different benchmarks. Currently, the handout has been released with the course update, and there are no further developments yet.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

DeepSeek releases DeepSeek-V4.1-Flash-1017

On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, which the company says delivers stronger benchmark performance, faster task completion, and shorter outputs, achieving a Pareto improvement in performance and cost; it is now available on the official website, App, and API, and the model also launched on OpenRouter the same day. The company also announced that the Search and Fetch APIs are open to everyone, priced at $0.75 per 1,000 calls per endpoint, with calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex temporarily free, and billing starting at 16:00 UTC on October 31, 2026. On October 17, Thomas Wolf commented that DeepSeek called this a "minor update," yet Terminal-Bench 4.0 rose from 31.2 to 42.3. On October 19, Arena announced that the model, with a score of 1670, became the highest-ranked open-source model on Code Arena: WebDev. On October 20, OpenCode updated it to version 1017 in Go and Free; Unsloth released a GGUF that can run locally, with the 1-bit quantized version retaining about 88.2% accuracy and shrinking in size from 508GB to 250GB; Arena then announced that it topped the Agent Arena open-source model leaderboard.

Meta launches Muse Wonder and Muse Wonder Full models

On October 13, 2026, Meta released the Muse Wonder series, comprising two models, Muse Wonder and Muse Wonder Full, which it called its most capable models to date, achieving SOTA on personal and professional agent tasks, covering professional domains such as law, finance, and health as well as frontier coding capabilities. Both models feature Active Memory and Active Focus, and were trained, evaluated, and released according to the Meta Superintelligence Scaling Framework. According to Artificial Analysis data, the two models scored 52 and 56 respectively on Intelligence Index v4.3 at maximum reasoning effort, higher than the 48 of Muse Spark 1.3 (max); Muse Wonder is already available on Meta's first-party API and Muse Code, while Muse Wonder Full is available only to Muse Code users. On the same day, Arena added the two models to Agent Arena testing, and Simon Willison released the llm-meta-ai 0.2 plugin supporting the new models. On October 14, Arena announced Muse W

English translation is being prepared.