TLDR IT 2026-08-31

Anthropic says it will permanently raise Claude Code's standard weekly limits by 25% for Pro, Max, Team, and seat-based Enterprise plans starting September 14. However, because users currently have a temporary 50% boost, the change will reduce their available usage by about 17% compared with today. Anthropic says the temporary increase remains through September 13.

Unexpected enterprise demand for local AI compute is reportedly driving Mac mini and Mac Studio sales hard enough that Apple accelerated its latest hardware launch. Companies are clustering Mac Studios to run large models, while constrained supply has pushed some buyers toward alternatives such as Nvidia's DGX Spark.

Major cloud-software availability incidents arise from persistent, fundamentally unavoidable risks rather than isolated mistakes. The main risk areas are resource saturation, networking failures in distributed systems, security mechanisms that block legitimate users, and uncommon operational changes that require less-practiced tools or human interventions.

Owner.com says its own customer research initially showed restaurant owners were wary of AI, but that conclusion was outdated within three months. CEO Adam Guild walks through how the company spent three years rebuilding around AI, including changing its product, workflows, and internal operating model, a useful look at what an actual AI transformation requires beyond simply adding AI features.

Intel laid out three architectures spanning data center to endpoint AI: the 256-core Diamond Rapids Xeon, Crescent Island inference GPU, and already-shipping Wildcat Lake client SoC. The latter's 17-TOPS NPU notably falls short of Microsoft's 40+ TOPS Copilot+ requirement, while Diamond Rapids isn't expected until 2027.

SkillRepo has launched Skillsets, a governance feature that lets teams assign graded skill collections to groups of repositories and control which agent instructions they run. It adds approval queues for external updates, repository compliance and drift visibility, sync records, and rollback capabilities, while preserving the open skill format and existing session-start delivery process.

Cisco is adding Supermicro's liquid- and air-cooled rack-scale systems to its Secure AI Factory with NVIDIA, wrapping them with Cisco networking, validation, support and lifecycle services. Systems are expected through Cisco's enterprise and channel ecosystem beginning in October, with architectures designed to scale from roughly 1,000 to more than 100,000 GPUs.

Kafka could support elastic partitioning, dynamically splitting and merging key ranges to scale ingestion and processing while balancing load. The design could preserve per-key ordering and Kafka's batch-oriented efficiency, but it would complicate consumer state locality, repartitioning, replay tracking, and recovery. Eliminating partitions entirely would sacrifice important batching and storage benefits.

Glean argued that enterprise AI reliability depends primarily on supplying agents with relevant company context, rather than adding more oversight after failures occur. The company introduced Glean Tau, a desktop-first product that combined local files and workflows with its cloud context graph, and reported that Glean Assistant outperformed Claude Cowork on cost and preferred outputs in an internal benchmark.

The experiment evaluates Drasi's standing Cypher queries against a live AKS Kubernetes resource graph to detect persistent, cross-resource drift.

A $19.4 billion New Jersey AI data center is facing complaints over gas turbines, construction, and noise, highlighting how power and permitting are becoming real constraints on scaling AI infrastructure.