Search “Search”

Anthropic shares when to fold subagents into the main agent loop

In comparisons across multiple enterprise deployments, Anthropic's Applied AI team found that a single agent with skills often outperforms subagent architectures in quality, with lower per-task cost and latency, and presents a migration method for folding subagents into the main loop. The article notes that subagents remain suitable for scenarios where context bloat degrades performance, independent tasks need to be processed in parallel, or tool selection conflicts with instructions, and recommends first treating subagents as tool calls, then converting them one by one into skills and MCP tools, using end-to-end evaluations to compare performance before and after. The article mentions that Claude Code's system prompt for more advanced models has been reduced by over 80%, accompanied by Claude API skills such as build-eval, prompt-audit, hillclimb, and cost-optimize.

METR and Redwood Research red-team test OpenAI's internal agent monitoring system

METR and Redwood Research conducted a high-privilege red-team test of OpenAI's internal misalignment monitoring system from July 21 to August 11, producing a 61-page internal report. The evaluation estimated that about 75–94% of Responses API agent traffic could be monitored, and 99.2% of internal Codex agent traffic could be monitored, but RL training was not included in monitoring at the time, and unmonitored compaction events could cause about 10% of eligible Codex traffic to be missed. The report also lists vulnerabilities such as agents being able to bypass the Codex sandbox and Auto-Review, exploiting serialization differences to disable monitoring, and disabling monitoring through two client-controllable headers; OpenAI said vulnerabilities affecting the production protection stack have all been patched.

OpenCode 2.1 Released: One Coding Agent Supports Multiple Models

The open-source AI coding agent OpenCode has released version 2.1, with changes to the SDK, HTTP messages, events, and plugin interfaces. Vendor-related data has been moved under tool.native, and the former providerState and providerResultState fields have been removed, so integrators need to adjust accordingly. Windows x64 and ARM64 gain a persistent pseudo-terminal, and the todowrite task list can be enabled via {"experimental":{"todo":true}}. This version builds on 2.0, which unified the terminal, web, and desktop clients onto a single local backend service.

Fortinet Issues Emergency Update Due to AI-Discovered Critical Vulnerabilities

Fortinet announced unscheduled security updates for FortiOS, FortiManager, FortiAnalyzer, FortiMail, and FortiAuthenticator to fix critical and high-risk vulnerabilities discovered by the Frontier-AI model during security testing, and urged customers to install them as soon as possible. Affected versions include FortiOS 8.0.2, 7.6.8, 7.4.13, 7.2.14, and others; FortiSASE will update automatically by this Thursday. Fortinet also warned that similar AI capabilities could be used by attackers to reverse-engineer patches and construct working exploits within hours or days.

Mave Completes $28 Million Series A Funding, Bringing Total Funding to $34 Million to Advance Agentic SecOps

Mave announced a total of $34 million in funding, including a $28 million Series A led by Red Dot Capital Partners, with the funds to be used to advance AI research, strengthen enterprise products, and expand into international markets. Its agentic security operations platform connects to enterprises' existing tools via API with read-only permissions, combining fine-tuned small language models with frontier large language models to complete investigations and approved responses within boundaries set by security teams. Mave says the platform is already running in more than 200 security operations centers, with ARR growing more than 400% over the past year, though it did not disclose absolute figures.

Hugging Face Releases Open-Source Model Carbon-X, Which Can Analyze and Generate DNA Sequences Using Natural Language

Hugging Face has released the open-source model Carbon-X, combining DNA analysis and generation with natural language so that researchers can ask questions directly and receive predictions and explanations. The model is based on Qwen3-14B and was trained on genomic sequences and natural language from humans, other mammals, plants, and fungi, with viruses explicitly excluded from the training data. Researchers can run it locally, avoiding the sharing of sensitive patient data with external AI companies, and can also generate new DNA sequences to assist gene therapy research.

MiniMax Code Desktop Launches Mini App

MiniMax has launched Mini App in local sessions of the MiniMax Code desktop app (macOS / Windows). It is a persistent application with a frontend and a Node.js backend, installed as a plugin, that can call and be called by the Agent, keeping work on a continuously alive page rather than scattered across conversations. This release includes the Mini App runtime (persistent data directory, hosted MCP, Agent API, Host Connectors, and in-page tool approval), /miniapp-creator for generating a page with a single sentence, and the official Mini Apps in the plugin marketplace and the open-source community repository MiniMax-Code-MiniApps. According to the official announcement, account credentials are kept by the host and cannot be accessed by plugins or pages, and content passed from a page to the Agent first enters the input box for review.

Stanford CS 329Z Lecture 9 Outlines the Selection of Agent System Optimization Methods

Lecture 9 of Stanford's CS 329Z course presents a selection diagram for Agent system optimization: the prerequisite for optimization is having data, a system, and evaluation metrics all in place; generally, context optimization (prompts and few-shots) is done first because it is cheap, fast, and sample-efficient; when context optimization is insufficient, or when you want to distill capabilities into a small model, or when prompts grow longer and longer, then you move on to changing the weights. The lecture notes also review reflective prompt optimization methods such as OPRO, MIPROv2, TextGrad, and GEPA; weight optimization paths such as SFT, RL (PPO vs. GRPO), and on-policy distillation; and approaches that let agents self-optimize, such as the Darwin Gödel Machine, Meta-Harness, and karpathy's autoresearch. They also point out that agents are already decent optimizers, making evaluation metrics the link that most needs to be controlled.

Anthropic releases personal agent Claude Duet, built on the Opus 5.5 model

Anthropic has released a personal agent app called Claude Duet, positioned as a "proactive agent" that can run in the background and proactively push tasks to users; it is currently in research preview with a waitlist. The tool is built on the flagship Opus 5.5 model released at the end of September, and it pushes a daily morning briefing from connected apps, notifies users when something needs attention, and requires user approval for some tasks. Mike Krieger, co-lead of the Anthropic Labs team, says it can judge when to act proactively; Karan Ahuja, who leads Duet development, previously headed product management at Meta Superintelligence Labs.

Claude Code v2.1.307 released

Claude Code v2.1.307 adds the `/claude-api api-version` command, which can migrate Claude API code to API version `2026-10-21`, and adds a `claude_code.safety_event` OpenTelemetry event for each tool call that is blocked, prompted, or returned by permission rules, auto mode, safety checks, hooks, or the sandbox. This version also fixes the issue where alias models such as `opus[1m]` reported model-not-found when `CLAUDE_CODE_DISABLE_1M_CONTEXT` was set, as well as the issue where MCP servers in `claude -p` falsely reported "authentication required" when the macOS keychain was locked.

Anthropic Launches Proactive Agent App Claude Duet

Anthropic has opened a research preview of Claude Duet, a standalone app that works as a proactive agent, handling tasks across connected apps, tracking email and calendar, and continuing to run in the cloud after the app is closed. The feature costs nothing extra but counts toward existing plan usage, and background work may cause users to hit limits sooner; it is currently invite-only and not yet available on Team and Enterprise plans. Claude Duet is powered by Claude Opus 5.5 and is available on iOS, Android, and the web, supporting Stripe unpaid invoice checks, Link and Shop Pay payments after user approval, as well as plugins and a cloud browser.

Will Larson on the revised rules of engineering management: How AI is changing engineering management and the engineer-to-CTO path

Imprint CTO Will Larson discussed on a podcast how AI and shifting industry expectations are reshaping engineering management, the career ladder, and the growth path from engineer to CTO, and advised middle managers to return to frontline management and relearn technical skills. Drawing on his experience at Digg, Yahoo, Stripe, Uber, Carta, and Calm, he discussed cases such as the Digg v4 rewrite and how Yahoo's server approval process spawned a "black market," as well as how AI is bringing technical decision-making power back to management.

How VocabX cuts real-world inference costs by 77% with specialized SLM vocabularies

VocabX selects tokens by cost rather than frequency and simultaneously rebuilds subword segmentation and multi-word merging, compressing the vocabulary of domain-specific small models to a smaller scale and reducing inference costs by 77% in real-world applications. Taking a scoring model as an example, generating one token costs 7.4 times as much as reading one token, and the output-layer vocabulary shrinks from about 152,000 rows to about 65,000 rows. This method targets the memory bandwidth bottleneck of small models such as Qwen3-4B at medium batch sizes, whereas methods such as FVT, MWT, and in-place BPE select tokens only by frequency and address only a single level.

Sarvam AI releases enterprise agent knowledge fabric architecture and benchmark results

Sarvam AI introduced a knowledge fabric built for enterprise agents, composed of two layers, a knowledge tree and a domain ontology, which compiles documents once into a structured layer that agents can navigate, query, and cite. On the Harvey LAB benchmark, answers using GLM 5.3 met 84.2% of expert grading criteria, higher than agentic RAG's 55.1%, with average task cost dropping from $0.64 to $0.28; on Sierra's tau-knowledge banking benchmark, it achieved accuracy comparable to the baseline with 64% fewer tokens. The knowledge tree is already used in the Sarvam cowork product, while the ontology is built domain by domain and enterprise customer by enterprise customer.

Mastra Launches Mastra Observability Managed Platform

Mastra has released the Mastra Observability managed platform, which can connect directly to Mastra framework projects, automatically trace agent runs, and view the latency, tokens, and cost of model and tool calls. The platform supports online scorers to monitor correctness, team annotation of traces to build evaluation datasets, and running evaluations in CI via Gates and Verdicts to catch regressions. It offers 100k observability events per month free with 15-day retention; beyond that, it costs $10 per 100k events.

Inception Releases Diffusion Language Model Mercury 3, Claiming Speeds Four to Eight Times Faster Than GPT-6 Luna

Inception has released the diffusion language model Mercury 3, claiming it wins 8 of 12 benchmarks and runs four to eight times faster than GPT-6 Luna and Claude Haiku 5.5. In Inception's own testing, Mercury 3 scored 76.84 on IFBench, 15 points higher than GPT-6 Luna; it scored 51.08 on SWE-Bench Atlas, versus 33.33 and 32.26 for its competitors. Claude Haiku 5.5 led on the DeepSWE long-code test and long-text retrieval. The model generates 1,068 tokens per second on Nvidia GPUs, is priced at $0.20 per million input tokens and $0.70 per million output tokens, with a 50% discount during the launch period, and is now available on the Inception platform as well as OpenRouter, Baseten, and Vercel; there are currently no independent tests.

Nvidia launches Open Agent Safety Platform, using BlueField-4 DPU to monitor agents

At GTC Berlin, Nvidia launched the Open Agent Safety Platform, adding hardware-level controls around the open-source runtime OpenShell; a BlueField-4 DPU works with Sentry software to independently monitor agent behavior and, when it crosses boundaries, isolate and stop it within milliseconds. OpenShell supports third-party agent frameworks such as Anthropic's Claude Code and OpenAI's Codex, and Nvidia says more than 100 organizations already use the platform's technology. Nvidia also introduced the Agent Knowledge Foundation and the Auto-Ontology library, designed to let agents access structured and unstructured data without enterprise data leaving local premises, and expanded its Nemotron 3.5 Super VL, Nemotron 3.5 Bolt 74B-A9B and Kumo series models.

AI Tries to Take Over Scanlation Groups' Jobs, Only to Find They Don't Rely on Translation to Make a Living

Models such as DeepSeek, GPT, and Qwen can already handle the collection, upscaling, translation, and typesetting of manga scanlation, but when they encounter explicit content or copyrighted content they refuse to translate outright, so a "human feedback" step is still needed to fill the gaps. The work of scanlation groups has therefore shifted to proofreading tone, handling puns and cultural references, adjusting for differences between Chinese and Japanese typesetting, and even drawing clothes onto exposed characters to pass review.

English translation is being prepared.

English translation is being prepared.