Insights & News
Ideas on websites, AI automation, digital marketing, AI news, and VMTS updates.
MODULE.001 //
All posts
Ideas on websites, AI automation, digital marketing, AI news, and VMTS updates.
AWS adds Bedrock AgentCore Runtime Instances for persistent production agent compute
AWS has added managed EC2-backed Runtime Instances to Bedrock AgentCore, with multi-agent runtimes, shared sessions of up to 14 days, GPUs, containers, and restart support.
Cloudflare AI Search connects websites, files, and MCP retrieval to agents
Cloudflare’s AI Search update makes it easier to index sites and structured or unstructured files, then expose public or Access-protected search and MCP endpoints per namespace.
ByteDance releases SeedRealtime, an audio-visual full-duplex LLM
ByteDance Seed says SeedRealtime continuously understands audio, video, and text, with proactive interaction and tool calling for more natural real-time experiences.
OpenAI says it cannot rule out critical cyber capability in Astra
OpenAI says preliminary internal evaluations of its upcoming Astra model show enough progress in agentic coding and cybersecurity that critical capability cannot yet be ruled out.
Microsoft Work IQ Developer Tools connect Copilot plugins to a full agent lifecycle
Microsoft’s Work IQ Developer Tools preview takes Microsoft 365 Copilot skills, connectors, and declarative agents through creation, validation, evaluation, publishing, and monitoring.
GitHub makes Copilot code review effort levels generally available
GitHub has made Lite and Balanced effort levels generally available for Copilot code review, letting teams match review depth to pull-request complexity and risk.
GitHub brings enterprise MCP allowlists to Copilot managed settings
GitHub now lets enterprise owners control which MCP servers Copilot can run through allowedMcpServers and deniedMcpServers policies, with URL, command, and fail-closed enforcement.
GitHub Copilot cloud agent adds a reasoning-level control
GitHub now lets Copilot cloud agent users choose a reasoning level for supported models, trading more capable work on complex tasks for higher token and credit use.
OpenAI and APA bring youth mental-health science into AI safety
OpenAI and the American Psychological Association plan to develop resources for families and practitioners while grounding youth AI safeguards in psychology and lived experience.
Agent Plugins 1.0.0 gives AI agents a portable package format
Google has joined the Agent Plugins 1.0.0 core maintainers, backing a portable directory format for Agent Skills, MCP servers, and client-specific extensions.
Anthropic reduces Fable 5 biology fallbacks while keeping dual-use limits
Anthropic says an updated Fable 5 biology classifier cuts biology-related fallbacks by about 85%, while dual-use virology, toxicology, and molecular-design requests remain restricted.
Microsoft makes the GitHub Copilot Agent stable inside Agent Framework
Microsoft’s stable .NET and Python integration brings the GitHub Copilot coding harness together with Agent Framework tools, approvals, MCP, streaming, and OpenTelemetry tracing.
Google Cloud API Gateway adds one OpenAI-compatible entry point for model routing
Google Cloud’s public preview lets applications send OpenAI-compatible requests through one gateway and route them to Gemini, Claude, or open-weight GPT models on Vertex AI.
AnyMind’s AnyAI Agent turns marketing data and campaign work into a governed loop
AnyMind’s new AnyAI Agent connects social, ecommerce, advertising, and enterprise data across an Observe, Think, Create, and Govern workflow.
GitHub Copilot automations let issue and pull request comments trigger agents
GitHub now lets issue and pull request comments trigger Copilot cloud agent automations for documentation, error investigation, and follow-up engineering work.
NVIDIA Alpamayo 2 Super brings an open commercial reasoning model to robotaxis
NVIDIA has made Alpamayo 2 Super available for commercial use under OpenMDW-1.1, with trajectories, causal traces, intent outputs, and 2D-grounded vision responses for autonomous driving.
QwenWork pushes Alibaba’s workplace agent from chat toward files, data, and scheduled delivery
Alibaba’s QwenWork is an all-in-one workplace AI agent platform connecting local files, cloud data, DingTalk, skills, connectors, and scheduled tasks around deliverable outputs.
Microsoft Research Orchard turns the agent environment into reusable infrastructure
Microsoft Research has released Orchard, an open framework that reuses environments, sandboxes, training pipelines, and evaluations across coding, browser, and personal-assistant agents.
Liquid AI’s LFM2.5-2.6B brings agent workflows to phones and local hardware
Liquid AI has released a 2.6B-parameter on-device model for planning, tool calling, and multi-step agent work, with weights available through Hugging Face.
Deepnote Agent Workspace turns shared data context into reusable workflows
Deepnote’s Agent Workspace lets data teams turn trusted analyses into skills, agents, and apps with shared permissions, integrations, schedules, and APIs.
MODULE.002 //
需要協助或查詢更多？
告訴我們您的網站、維護或推廣需要，我們會盡快回覆。