Microsoft Cuts 4,800 Jobs in 'AI-Driven' Restructuring — Xbox CEO: "The Business Is Not Healthy"
On July 6, Microsoft announced a major round of layoffs centered on its Xbox division. According to Bloomberg, Xbox will cut 3,200 jobs—about 20% of its total workforce—within the next 12 months. 1,600 employees are affected immediately, with the remainder phased in over time. Xbox also announced the sale or spin-off of five game development studios.
In a letter to employees, Xbox CEO Asha Sharma said "our business is not healthy today," emphasizing that the current state—margins 3 to 10 times lower than comparable businesses—"needs a reset." Reuters also reported the cuts as part of an "AI-driven wave of tech industry layoffs."
In addition, a separate round of 3,200 layoffs is underway at Microsoft outside of Xbox, centered on the sales division. Chief People Officer Amy Coleman explained that "how we build products and what customers want are changing." AI-driven efficiency gains appear to be behind the job cuts, and this looks like a symbolic case of Big Tech's broader "AI pivot" starting to hit employment directly.
US Treasury Warns of "AI Bubble" Risks in Internal Report
It has come to light that the US Department of the Treasury possesses an internal report warning of the systemic risks posed by the overheating AI industry. The news was reported by the nonprofit media outlet Notus.
The detailed contents of the report remain undisclosed, but it is believed to discuss the surge in AI investment, the validity of valuations, and spillover risks to the financial system. The very fact that the Treasury is conducting this kind of internal analysis suggests that government economic authorities have begun seriously examining the sustainability of the AI boom.
Private investors and analysts have long pointed to the possibility of an "AI bubble," but reports that the federal agency in charge of economic policy has been cataloging these risks in internal documents are likely to change the weight of the debate.
Meta's AI Glasses Get Restrictions "That Would Make Comcast Proud"
Meta's AI-powered smart glasses (Ray-Ban Meta) have been hit with new restrictions that are drawing a flood of user complaints. Technology criticism site TechDirt mocked them as "ridiculous usage restrictions that would make Comcast (the giant US cable TV provider) proud."
Details on exactly what restrictions were added remain limited based on the information gathered. The topic has, however, drawn 6 points of interest on Hacker News, where the community has shared concerns that it could become "a major barrier to AI glasses adoption."
Meta positions its AI glasses as the next-generation computing platform, but if feature restrictions are designed in a way that undermines the user experience, they could well become a drag on adoption.
"It's Just AI Talking to AI" — AI Communications That Damage Workplace Relationships
Fortune reported on how AI use among corporate middle management is damaging human relationships. The title quotes a worker saying: "At the end of the day, it's just his AI talking to my AI."
The article points out that when managers delegate communication to AI tools, it hinders the relationship-building and trust formation that would otherwise develop naturally. It highlights a structural problem: communications handed off to AI in the name of efficiency ultimately weaken organizational cohesion.
As AI tools rapidly make their way into the workplace, designing the right balance between "efficiency" and "human connection" has become a new challenge for management.
llama.cpp ARM Optimization Delivers 5x Speedup — Leveraging LUTs for NVFP4 Dot Products
A PR optimizing NVFP4 dot product computation on ARM processors has been submitted to llama.cpp, a project drawing significant attention in the local LLM community. It leverages UE4M3 lookup tables (LUTs), achieving roughly a 5x throughput improvement in benchmarks with the Qwen3.5-4B-NVFP4 model—up from 1.89 t/s to 9.97 t/s.
The PR extends the UE4M3 LUT optimization previously introduced for x86 environments to an ARM implementation. Sharing the lookup table infrastructure between ARM and x86 also improves code consistency. According to comments from the developer, inference with small NVFP4 models on CPU-only environments is approaching practical levels.
Improvements to the foundational technology that governs local LLM inference efficiency represent major progress for AI usage in environments with limited GPU resources.
GitHub and Others Accused of "Steering" Opinion on California's FOSS License Bill
The Software Freedom Conservancy (SFC) has criticized generative AI companies such as GitHub for engaging in deliberately misleading lobbying around a California bill concerning free and open source software (FOSS) licenses.
According to an SFC blog post, these companies are distorting the bill's contents and working to make ordinary developers mistakenly believe that "their own rights are under threat." For generative AI companies, the use of open source code as training data is an area where significant interests are at stake, and the direction of license legislation directly affects their business models.
The legal and ethical conflict between the open source community and AI companies over code usage rights continues, and the fate of this bill looks likely to serve as one indicator of the balance of power between the two sides.
Of 1,397 AI Agents Being Tracked, 43 Prominent Projects Have "Vanished"
AgentCrush, a platform that tracks AI agents, has published a report showing that of the 1,397 AI agent projects it monitors, 43 prominent projects (a combined 400,000 GitHub Stars) are in a "Ghost" state—no longer active.
The AI agent space saw an explosive increase in projects across 2025–2026, but this data highlights the reality that, contrary to the apparent boom, many projects are not sustaining themselves. Even projects with high Star counts frequently see development stall, and the transition from "launching on the hype" to "sustainable development" has become an industry-wide challenge.
---