TLDR IT 2026-08-11

Buying AI licenses is the easy part. Getting people to use them every day is the real work.

Join Boston Scientific and BrainStorm on August 27th at 12pm Mountain/2pm Eastern to hear how one enterprise closed that gap by bringing HR and IT together and giving employees an AI skills ladder they can actually climb.

By the end of the session, you'll have learned:

→ A practical framework to manage AI adoption as a behavior problem, and

→ An honest account of how one enterprise is working through it right now.

Attend live to join the discussion →

Gartner expects AI-optimized IaaS spending to grow 96% this year to $42 billion, with inference overtaking training as the largest source of demand. Agentic AI and production deployments are pushing enterprises toward significantly more continuous compute consumption.

Meta Muse Glimmer is a 30-billion-parameter open-weight model optimized for autonomous agents and designed to run on consumer hardware. The Apache 2.0 release marks a renewed push by Meta around open AI models.

The Django Python web-framework project has accepted a proposal to replace its complex mix of short-lived feature and long-term-support releases with an annual release cycle. Beginning with the expected January 2028 release, every feature release will receive three years of support.

Shipping Device Bound Session Credentials at Report URI and open-sourcing the server-side implementation exposed several browser and concurrency pitfalls. Refresh flows correctly begin with a 403, session-state clobbering can silently disable binding, cookie and challenge rotation are required, and legitimate overlapping requests can trigger false stolen-cookie detections.

Cybersecurity has substantially improved technical detection and containment while leaving enterprise incident response largely fragmented. The article advocates treating incident response as a coordinated command-center discipline, particularly as attackers' exploitation and intrusion timelines continue to shrink.

Longer context windows alone aren't enough for agents that operate across hours or days. The emerging infrastructure challenge is persistent memory that can decide what information to retain, retrieve, and discard.

Meta Muse Glimmer is a 30B open-weight dense model with a 120K-plus context window designed for local, multi-step AI agent workflows. NVIDIA says it can run on a single NVIDIA GPU at up to 20K tokens per second, keeping sensitive data on device, and provides deployment through NIM, vLLM, SGLang, Hugging Face weights, and NVIDIA platforms.

Cloudflare's Kitesurf is a cloud-hosted browser optimized for agents, stripping away human-facing browser overhead while reducing CPU, memory, and token costs. It runs on Workers and lets agents navigate sites, fill forms, and perform browser-based tasks programmatically.

External dependencies do not invalidate vertical slices. The key is to distinguish slices, modules, and bounded contexts rather than treating every feature as an autonomous context. In a transport-order workflow, task-specific slices use handlers with explicit dependencies, allowing them to call external systems or access shared data without collapsing back into generic status updates.

OpenAI is introducing $125/month Premium seats with 5x the usage of standard ChatGPT Business seats and no five-hour usage limit. Companies can mix seat types and centrally manage usage, billing, and spend.

Researchers are increasingly testing unreleased models with safeguards disabled, making the security of evaluation infrastructure itself a growing concern.

Brex describes an approach to securely deploying powerful agents by monitoring and restricting their network behavior rather than trying to anticipate every action in application code.

Conventional shadow-IT tools cannot detect much MCP sprawl because many MCP servers run locally over stdio rather than crossing a monitored network boundary.