Cloudflare Computer 0.7 makes agent code resumable and interruptible. The important catch: recovery replays code, and unrecorded external actions can be repeated even if they already succeeded.
Cloudflare discloses cross-tenant GLM-5.3 Workers AI responses caused by a KV-cache rejection bug; SGLang v0.5.22 is recommended for disaggregated inference operators.
A new Performance Max sourced-videos setting can add up to ten crawled videos per asset group, uploaded to a Google-owned branded YouTube channel. It is on by default for new campaigns, but sources require approval.
Agent skills join prompts and models as managed runtime configurations; approvals and pinned versions make changing agent behaviour auditable rather than an untracked edit to skill files.
Quattro’s unified programmable shell is a real architecture change rather than a theme refresh. Omarchy 4.0.2 now hardens package, installer, SSH and input paths, while current user reports of Quickshell crashes and a runaway-memory event illustrate the new central shell’s blast radius.
Approved apps can move from the standard 20%/25% non-recurring service-fee rates to 15%/20% for new/existing installs from September 30, before any applicable billing fee. Current enrollment is limited to developer account groups with at least $1 million in earnings over the previous 12 months.
The two August 28 changes move a common production-agent problem out of bespoke application code: builders can derive memory boundaries from authenticated JWT claims, enforce them with Cedar policy, and organize the stored memory using runtime tenant dimensions.
Cloudflare Workflows now prices steps and persisted state on paid plans, making workflow structure and retention part of the cost calculation for durable jobs and AI automation.
A PHP patch emergency is now a live-incident-response story: examine internet-facing PHP exposure and suspicious activity back to at least October 17, then upgrade immediately.
A new npm granular-token scope lets CI stage package versions without permission to publish them, extending npm’s broader move toward least-privilege publishing after its install-script, trusted-publishing and malware-gate changes.
Microsoft Agent Framework PyPI versions 1.26.1 and 1.26.2 contained .pth-triggered credential stealers. Operators need to check resolved dependencies, isolate affected environments and rotate reachable secrets.
Search Console can now treat supported social and video accounts as properties, exposing Google impressions, clicks, queries and post-level performance for Instagram, TikTok, X and YouTube content.
Aurora Serverless now adds up to 16 ACUs per second at the initial scaling step, and eligible Aurora PostgreSQL clusters can scale as high as 384 ACUs.
The post-release evidence sharpens the original story. Qwen3.8-27B can retain useful agentic-coding performance at practical 4-bit sizes, but local model quality is not a property of the checkpoint alone: quantization, reasoning effort, context handling and the agent harness can materially change the result.
GLM-5.3-Flash combines open weights, multimodal coding/agent capability and an 18B-active sparse architecture with a large anonymous pre-launch trial. Z.ai has already issued a chat-template correction for early downloads, showing that day-one self-hosted deployments need artifact-level validation as well as model benchmarking.
Android Studio’s agent layer has crossed an important boundary from preview features into the stable channel: domain-specific skills are preloaded and auto-selected, while Gemma 4 can execute tool-calling code tasks locally without sending source code to a cloud model.
Microsoft’s Search automation has moved from a limited test into a default campaign-creation path. Advertisers can still switch AI Max off, but new Search campaigns now begin with a suite that can expand queries, creative and landing-page routing beyond the manually supplied keyword-and-ad structure.
The useful part of Smaug Agentic is not another frontier-style benchmark claim. Abacus.AI is publishing a drop-in Kimi K3 derivative that targets a specific production failure mode in coding agents: long runs that burn the reasoning budget without converging. The weights and model card are public, but the training data is not disclosed and the benchmark gains remain vendor-produced.