Search “Agents”

ops-agent releases v0.1.0-rc.1 source code release candidate

The Ops Agent project ops-agent has released its first release candidate, v0.1.0-rc.1, with commit d897d2c95e70, running on the Python 3.9+ standard library, targeting macOS/Linux. This version includes 309 built-in self-check items and 223 independent regression items; the release verification matrix passed across four combinations of Ubuntu/macOS × Python 3.9/3.12, and the W8 local drill passed 10/10. External acceptance, including real SSH, real model services, notification channels, and 7 consecutive days of post-fix observation, has not yet been completed.

View story:ops-agent releases v0.1.0-rc.1 source code release candidate

Shao Meng shares Claude blog: when to fold sub-agents into the main loop

On October 25, 2026, Shao Meng shared a Claude blog post's view on when to fold sub-agents into the main loop when building Agents. The blog argues that on the Claude Platform, a single main loop plus on-demand loading of Skills/tools often yields better quality, lower cost, and lower latency across multiple internal comparisons; one sub-agent per team plus aggregate orchestration often replicates the organizational chart, bringing problems such as context loss, Frankenprompts, and localized evals, and when one customer compared a multi-agent harness with a single loop, the latter was about 2× faster. The blog suggests reserving sub-agents only for scenarios such as parallelism, context bloat, or conflicts between tools and instructions, and offers a migration path of first building end-to-end evals and folding one at a time.

Shao Meng shares Claude blog: when to fold sub-agents into the main loop

On October 25, 2026, Shao Meng shared a Claude blog post's view on when to fold sub-agents into the main loop when building Agents. The blog argues that on the Claude Platform, a single main loop plus on-demand loading of Skills/tools often yields better quality, lower cost, and lower latency across multiple internal comparisons; one sub-agent per team plus aggregate orchestration often replicates the organizational chart, bringing problems such as context loss, Frankenprompts, and localized evals, and when one customer compared a multi-agent harness with a single loop, the latter was about 2× faster. The blog suggests reserving sub-agents only for scenarios such as parallelism, context bloat, or conflicts between tools and instructions, and offers a migration path of first building end-to-end evals and folding one at a time.

View story:Shao Meng shares Claude blog: when to fold sub-agents into the main loop

dashi-ppt-skill provides a PPT skill package for AI Agents

The open-source project dashi-ppt-skill is released as an Agent skill package, providing AI Agents with the ability to generate presentations. According to a report by Alibaba Cloud Developer Community AI, this skill package generates HTML presentations with a built-in editing console, and can export editable PPTX that restores each node when needed. The project includes 12 visual themes and 1,020 layout pages, following a template orchestration approach, and by default only replaces text without modifying the layout; the official claim is that it provides 8,576 adjustable controls. The project uses the AGPL-3.0 license and can be installed into local Agents such as Claude Code, Codex, and Cursor via npx skills add; exporting PPTX/PDF requires Chrome, Chromium, or Edge installed on the local machine. Currently there is only this one report, and no subsequent developments have been reported.

dashi-ppt-skill provides a PPT skill package for AI Agents

The open-source project dashi-ppt-skill is released as an Agent skill package, providing AI Agents with the ability to generate presentations. According to a report by Alibaba Cloud Developer Community AI, this skill package generates HTML presentations with a built-in editing console, and can export editable PPTX that restores each node when needed. The project includes 12 visual themes and 1,020 layout pages, following a template orchestration approach, and by default only replaces text without modifying the layout; the official claim is that it provides 8,576 adjustable controls. The project uses the AGPL-3.0 license and can be installed into local Agents such as Claude Code, Codex, and Cursor via npx skills add; exporting PPTX/PDF requires Chrome, Chromium, or Edge installed on the local machine. Currently there is only this one report, and no subsequent developments have been reported.

View story:dashi-ppt-skill provides a PPT skill package for AI Agents

Shao Meng introduces Tibo's nine-step process for building an AI UGC creator team with Agents

On October 24, 2026, Shao Meng compiled and introduced a process from Tibo: using Agents to build your own AI UGC creator matrix. The process is designed by buyer segment, with 3–4 AI creators who have fixed faces and daily routines, completing each day the tasks of finding viral hits, rewriting them in the character's voice, rendering with the same face, and scheduling 24 hours in advance; humans only review the queue and weekly topic updates. The process has 9 steps in total and comes with 12 prompts, starting from connecting to the Revid MCP and writing the brain.md master brain, through creating faces, setting topics, breaking down viral structures, running daily batches, operating independent accounts, and weekly reviews. The article emphasizes that account bios should state AI creator, that fake reviews and health/medical claims are prohibited, that the product should occupy only one beat of the video, and that content is only scheduled rather than published directly so as to leave room for human review. There are currently no subsequent developments.

AMD Unveils CPU-Centric Helion Rack Design for Agentic AI

At OCP Global Summit, AMD announced Helion, a CPU-centric AI data center compute infrastructure design for running large numbers of AI agents simultaneously. Unlike Helios, which stacks large numbers of GPUs, this design targets the CPU demands created by agents repeatedly executing programs and accessing databases. A configuration for large data center operators can hold up to 32,768 sixth-generation EPYC Venice CPU cores in a single rack, with model inference requests handed off to separate GPU systems; the enterprise configuration places EPYC CPUs and Instinct MI350P GPUs in the same rack, holding up to 16 GPUs, and integrates a token router that routes requests to high-performance models or self-hosted open-source models to lower AI operating costs. AMD will offer Helion as an open reference design and contribute the specification to OCP; eight server vendors including Cisco, Dell, HPE, Lenovo and Supermicro are developing related systems, with launch planned for November.

Arena ranks Kimi K3.1 Pro (Max) No. 11 in Agent Arena

On October 24, 2026, Arena added Kimi K3.1 Pro (Max) to the Agent Arena rankings, placing it at No. 11. The ranking is based on 4K multi-real-agent sessions, achieving a +7.6% net improvement, up 6 percentage points from Kimi K3 (Max)'s No. 25. Its median cost per task is $1.03, 13% lower than the previous version; it ranks No. 8, No. 9, and No. 10 in the Work, Code, and Chat categories, respectively. Arena said the model's open weights will be released soon, at which point it will become the top-ranked open-source model in Agent Arena, surpassing DeepSeek-v4.1-Flash. There are currently no subsequent developments.

Arena ranks Kimi K3.1 Pro (Max) at No. 11 in Agent Arena

On October 24, 2026, Arena added Kimi K3.1 Pro (Max) to the Agent Arena rankings, placing it at No. 11. The ranking is based on 4K multi-real-agent sessions, achieving a +7.6% net improvement, up 6 percentage points from Kimi K3 (Max)'s No. 25. Its median cost per task is $1.03, 13% lower than the previous version; it ranks No. 8, No. 9, and No. 10 in the Work, Code, and Chat categories, respectively. Arena said the model's open weights will be released soon, at which point it will become the top-ranked open-source model in Agent Arena, surpassing DeepSeek-v4.1-Flash. There are currently no subsequent developments.

View story:Arena ranks Kimi K3.1 Pro (Max) No. 11 in Agent Arena

OpenBMB Releases Research on Device-Cloud Collaborative Long-Horizon Agent Execution System

On October 24, 2026, OpenBMB released research on a device-cloud collaborative long-horizon agent execution system. This research, conducted by ModelBest in collaboration with Tsinghua University, uses EC2-Bench to evaluate four device-cloud collaboration strategies on 128 executable tasks, simultaneously measuring task completion, resource cost, and privacy exposure. The results show that with on-device Qwen3.5-9B plus cloud-based GLM-5.3, the best collaboration score is 60.46, about 73% of the pure-cloud score of 82.70, at a cost of only $8.73, about 15% of the pure-cloud cost of $56.42. The research also found that weak on-device models are suited to having an external router let the cloud execute, with the 2B model scoring 39.97 versus 15.86; strong on-device models are suited to the cloud providing only suggestions, with the 9B model scoring 60.46. The research has now been made public, with no subsequent progress yet.

OpenBMB Releases Research on Device-Cloud Collaborative Long-Horizon Agent Execution System

On October 24, 2026, OpenBMB released research on a device-cloud collaborative long-horizon agent execution system. This research, conducted by ModelBest in collaboration with Tsinghua University, uses EC2-Bench to evaluate four device-cloud collaboration strategies on 128 executable tasks, simultaneously measuring task completion, resource cost, and privacy exposure. The results show that with on-device Qwen3.5-9B plus cloud-based GLM-5.3, the best collaboration score is 60.46, about 73% of the pure-cloud score of 82.70, at a cost of only $8.73, about 15% of the pure-cloud cost of $56.42. The research also found that weak on-device models are suited to having an external router let the cloud execute, with the 2B model scoring 39.97 versus 15.86; strong on-device models are suited to the cloud providing only suggestions, with the 9B model scoring 60.46. The research has now been made public, with no subsequent progress yet.

View story:OpenBMB Releases Research on Device-Cloud Collaborative Long-Horizon Agent Execution System

Anthropic open-sources OS-level Agent sandbox sandbox-runtime

Anthropic has open-sourced sandbox-runtime (srt), providing AI agents with container-free OS-level isolation that restricts the files and network hosts processes can access. According to a MarkTechPost report on October 24, 2026, the npm package @anthropic-ai/sandbox-runtime v0.0.81 is licensed under Apache-2.0 and requires Node.js 22.12+; macOS uses Seatbelt, Linux relies on bubblewrap, socat, and ripgrep, and Windows support is in alpha. Network access is denied by default, with traffic allowed through a host-side proxy based on a domain allowlist. Anthropic says permission prompts dropped 84% during internal Claude Code use, a self-reported figure. There are currently no subsequent developments.

Anthropic open-sources OS-level Agent sandbox sandbox-runtime

Anthropic has open-sourced sandbox-runtime (srt), providing AI agents with container-free OS-level isolation that restricts the files and network hosts processes can access. According to a MarkTechPost report on October 24, 2026, the npm package @anthropic-ai/sandbox-runtime v0.0.81 is licensed under Apache-2.0 and requires Node.js 22.12+; macOS uses Seatbelt, Linux relies on bubblewrap, socat, and ripgrep, and Windows support is in alpha. Network access is denied by default, with traffic allowed through a host-side proxy based on a domain allowlist. Anthropic says permission prompts dropped 84% during internal Claude Code use, a self-reported figure. There are currently no subsequent developments.

View story:Anthropic open-sources OS-level Agent sandbox sandbox-runtime

Agent Flow Designer provides visual agent workflow orchestration capabilities

On October 24, 2026, the Tencent Cloud Developer Community published an article introducing Agent Flow Designer (AFD) v2.2.2 as a visual agent workflow orchestration plugin for the QwenPaw platform, offering the ability to drag-and-drop build, publish, deploy, and manage multi-node Agent workflows. The plugin is based on the AntV X6 graph editing engine and provides 6 types of nodes: Start, Agent, Tool, Routing Decision, End, and Ask. Routing Decision supports three modes: rule matching, LLM intelligent routing, and hybrid mode, and it features version snapshot publishing, one-click rollback, and a /run external execution interface. The article states that it is compatible with QwenPaw / QzAgent ≥ 2.0.0, and that workflow data is persisted via atomic writes to JSON files. There are currently no subsequent development facts.

View story:Agent Flow Designer provides visual workflow orchestration for QwenPaw

Shao Meng introduces Tibo's nine-step process for building an AI UGC creator team with Agents

On October 24, 2026, Shao Meng compiled and introduced a process from Tibo: using Agents to build your own AI UGC creator matrix. The process is designed by buyer segment, with 3–4 AI creators who have fixed faces and daily routines, completing each day the tasks of finding viral hits, rewriting them in the character's voice, rendering with the same face, and scheduling 24 hours in advance; humans only review the queue and weekly topic updates. The process has 9 steps in total and comes with 12 prompts, starting from connecting to the Revid MCP and writing the brain.md master brain, through creating faces, setting topics, breaking down viral structures, running daily batches, operating independent accounts, and weekly reviews. The article emphasizes that account bios should state AI creator, that fake reviews and health/medical claims are prohibited, that the product should occupy only one beat of the video, and that content is only scheduled rather than published directly so as to leave room for human review. There are currently no subsequent developments.

View story:Shao Meng introduces Tibo's nine-step process for building an AI UGC creator team with Agents

Artificial Analysis publishes AA-AgentPerf results for Muse Spark Open

Artificial Analysis on October 24, 2026 published on AA-AgentPerf preliminary test results for Meta's newly open-weight model Muse Spark Open. The model is a 816B-parameter, 42B-active MoE, scoring 39 on the Artificial Analysis Intelligence Index. Tests show that NVIDIA B300 and GB300 each reach about 150M tokens/s per megawatt of power under throughput-first configurations; GB300 reaches about 108M and B300 about 92M at 100 tok/s output speed; AMD MI355X reaches 70M tokens/s per megawatt at 65 tok/s. Of the model's 62 layers, 46 use 2,048-token sliding window attention, reducing KV cache requirements at 1M context by about 75% versus full attention. Meta released an NVFP4 checkpoint, and AMD released an MXFP4 checkpoint. There are currently no subsequent development facts.

Artificial Analysis launches AA-AgentPerf results for Muse Spark Open

Artificial Analysis on October 24, 2026 published on AA-AgentPerf preliminary test results for Meta's newly open-weight model Muse Spark Open. The model is a 816B-parameter, 42B-active MoE, scoring 39 on the Artificial Analysis Intelligence Index. Tests show that NVIDIA B300 and GB300 each reach about 150M tokens/s per megawatt of power under throughput-first configurations; GB300 reaches about 108M and B300 about 92M at 100 tok/s output speed; AMD MI355X reaches 70M tokens/s per megawatt at 65 tok/s. Of the model's 62 layers, 46 use 2,048-token sliding window attention, reducing KV cache requirements at 1M context by about 75% versus full attention. Meta released an NVFP4 checkpoint, and AMD released an MXFP4 checkpoint. There are currently no subsequent development facts.

View story:Artificial Analysis publishes AA-AgentPerf results for Muse Spark Open

Bayer Crop Science builds an agentic decision engine based on AgentCore

Bayer Crop Science built ADE (Agentic Decision Engine), a multi-agent natural language interface based on Amazon Bedrock AgentCore, covering its Connected Decision Engine. According to information released by AWS Machine Learning on October 24, 2026, ADE runs 22 specialized agents in production, routed by a Supervisor Agent according to intent, and exposes backend APIs as MCP tools via AgentCore Gateway. In the first phase, about 80% of user questions were resolved without engineer intervention; fault diagnosis time was shortened from manual troubleshooting that previously took more than 1 hour to within 10 minutes. There are currently no subsequent development facts.

Bayer Crop Science builds an agent decision engine based on AgentCore

Bayer Crop Science built ADE (Agentic Decision Engine), a multi-agent natural language interface based on Amazon Bedrock AgentCore, covering its Connected Decision Engine. According to information released by AWS Machine Learning on October 24, 2026, ADE runs 22 specialized agents in production, routed by a Supervisor Agent according to intent, and exposes backend APIs as MCP tools via AgentCore Gateway. In the first phase, about 80% of user questions were resolved without engineer intervention; fault diagnosis time was shortened from manual troubleshooting that previously took more than 1 hour to within 10 minutes. There are currently no subsequent development facts.

View story:Bayer Crop Science builds an agentic decision engine based on AgentCore

Agent Flow Designer provides visual workflow orchestration for QwenPaw

On October 24, 2026, the Tencent Cloud Developer Community published an article introducing Agent Flow Designer (AFD) v2.2.2 as a visual agent workflow orchestration plugin for the QwenPaw platform, offering the ability to drag-and-drop build, publish, deploy, and manage multi-node Agent workflows. The plugin is based on the AntV X6 graph editing engine and provides 6 types of nodes: Start, Agent, Tool, Routing Decision, End, and Ask. Routing Decision supports three modes: rule matching, LLM intelligent routing, and hybrid mode, and it features version snapshot publishing, one-click rollback, and a /run external execution interface. The article states that it is compatible with QwenPaw / QzAgent ≥ 2.0.0, and that workflow data is persisted via atomic writes to JSON files. There are currently no subsequent development facts.