The written, source-linked counterpart to this video.
Published edition9 stories
01Updates
Google explains behavioral evaluation for coding agents
Google published a harness-engineering workflow that starts with dogfooding, adds local behavioral evaluations for discrete actions, and retains larger end-to-end suites for overall outcomes.
The Google Coding-Agent Harness Engineering page documents the Google update.
Agent builders can isolate which tool call or intermediate behavior regressed instead of relying only on a composite benchmark score.
Watch next
Check the next Google coding-agent harness engineering release note for rollout, availability, or quota changes.
Kimi Code CLI 0.42.0 makes Remote Control always on, adds read-only /btw side-agent tools, and adds permanent session deletion plus richer media attachment handling in the web client.
version 0.42.0 on September 9, 2026.
Coding-agent users gain a default remote-control path and a safer read-only side-agent tool surface without relying on an experimental environment flag.
Watch next
Check the next Kimi Code CLI 0.42.0 release note for rollout, availability, or quota changes.
Unity released a first-party Claude Code plugin that installs 29 engine skills, the Unity CLI, and Unity's MCP server for live Editor control.
The source records 29 skills.
Unity developers can give a coding agent engine-specific project guidance and live Editor tooling from a first-party package instead of relying on third-party instructions.
Watch next
Check the next Unity Plugin for Claude Code release note for rollout, availability, or quota changes.
CUDA Toolkit 13.4 Adds Windows on Arm Support and Greater Control over Shared GPUs
Every NVIDIA CUDA Toolkit release adds functionality and performance improvements that help developers get more from NVIDIA GPUs and the broader NVIDIA software.
The NVIDIA AI Blog release records 13.4 for NVIDIA.
Teams evaluating NVIDIA 13.4 can test how CUDA Toolkit 13.4 Adds Windows on Arm Support and Greater Control over Shared GPUs changes the documented workflow.
Watch next
Check the next NVIDIA 13.4 release note for rollout, availability, or quota changes.
1.17.0 (2026 09 09) Changed: Features feat: show manifest declared value statements on dashboard cards and rows by @hieptl in 16921 feat: Make automation run UI task outcome aware by @malhotra5 in 16937 feat: enable LLM provider connections on cloud by @juanmichelini in 16757 feat: allow editing a custom cron expression from the edit automation modal by @VascoSch92 in 16964 feat: unified cloud settings entrypoint for Agent Canvas by @jpelletier1 in 16883 feat: add local Agent Canvas Planner support by @VascoSch92 in 16178 feat(conversation panel): add conversation tags and filtering by @georgeglarson in 16600 feat: split automation permissions into view/manage with creator escape hatch by @tofarr in 17105 feat: display message timestamps on hover by @LosAveRoad in 17123 feat: support generic custom automation manifests in Canvas by @malhotra5 in 17129 Bug Fixes fix(chat): re derive.
The OpenHands Releases release records v1.17.0 / 1.17.0 for OpenHands.
OpenHands added control-plane detail for OpenHands updates v1.17.0; enterprise adoption depends on permissions, auditability, and data boundaries.
Watch next
For OpenHands, watch admin controls, audit logs, default permissions, and migration guidance before enterprise rollout.
Hugging Face updates IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license
As of September 8, 2026, the model is the top performing zero-shot model released under a permissive, commercial-friendly open-source license (Apache 2.0 and OpenMDW 1.0) among replicable, zero-shot models on the GIFT-Eval leaderboard. GIFT-Eval is a comprehensive time series forecasting benchmark designed to evaluate models across diverse forecasting scenarios; the model ranks #2 overall among replicable, zero-shot models.
The source records 2.0 / 1.0.
Teams evaluating Hugging Face 2.0 can compare its documented input scope, limits, and deployment fit.
Watch next
For Hugging Face, watch reproducible benchmark details, dataset coverage, and whether users can map the result to real workflows.
ci(desktop): exercise desktop packaging on a schedule; What's Changed ci(desktop): exercise desktop packaging on a schedule by @yiliang114 in #11519 fix(bridge): keep pending permission/question across refreshed session loads by @wenshao.
The Qwen Code Releases release records v0.3.0 for Qwen.
Qwen added control-plane detail for Qwen Code Desktop v0.3.0; enterprise adoption depends on permissions, auditability, and data boundaries.
Watch next
For Qwen, watch admin controls, audit logs, default permissions, and migration guidance before enterprise rollout.
LangChain updates Connections: managed credentials and per-caller identity for Managed Deep Agents
Learn how Connections in Managed Deep Agents securely manage credentials, support per user OAuth, and let agents act with each caller's identity.
The LangChain Blog page documents the LangChain update.
LangChain added control-plane detail for LangChain updates Connections: managed credentials and per-caller; enterprise adoption depends on permissions, auditability, and data boundaries.
Watch next
Check the next LangChain update for availability, pricing, limits, and proof that LangChain updates Connections: managed credentials and per-caller is being used.
If you administer GitHub Copilot Business or GitHub Copilot Enterprise, you can now centrally control which agent operations are blocked, require human approval
OFFICIAL | GitHub Changelog - github.blog · 2026-09-09
Google's September 9 developer guide recommends pairing broad end-to-end benchmarks with fast behavioral checks over intermediate agent actions to diagnose regressions
OFFICIAL | Google Coding-Agent Harness Engineering - developers.googleblog.com · 2026-09-09
Kimi Code CLI 0.42.0 removes the experimental flag for Remote Control, adds read-only tools to the /btw side agent, and expands web session and media controls
OFFICIAL | Kimi Code Changelog - moonshotai.github.io · 2026-09-09
Features Support GPT 6 Astra models 11869 Tool calling for goose agent 11627 Kotlin callers can now configure Databricks AI Gateway path 11631 Bug Fixes Require HTTPS for Snowflake connections 11745 Bind CLI session naming to trusted path 11787 Honor content capture opt out in tr
OFFICIAL | Goose Releases - github.com · 2026-09-08
Every NVIDIA CUDA Toolkit release adds functionality and performance improvements that help developers get more from NVIDIA GPUs and the broader NVIDIA software
OFFICIAL | NVIDIA AI Blog - developer.nvidia.com · 2026-09-09
A recap of August 2026 launches for AI builders across Amazon Bedrock, Amazon Bedrock AgentCore, and Strands: million token context for OpenAI models, cross Region inference, agents that run for up to 14 days on dedicated compute, expanded AWS GovCloud availability, and Strands R
OFFICIAL | AWS Machine Learning Blog - aws.amazon.com · 2026-09-09
Use Ollama models in ChatGPT Desktop Ollama models can now be used directly in ChatGPT Desktop, so you can keep your existing workflow while running open models.
OFFICIAL | Ollama Releases - github.com · 2026-09-10
DeepSeek V4.1 Flash is now available on AI Gateway with native image understanding.V4.1 Flash has vision support and accepts text and images in the same request
OFFICIAL | Vercel Changelog - vercel.com · 2026-09-09