Uber buys Glovo Kenya, BLUETTI BNPL solar, Honda kills US EV

Netflix's AI Platform team published a detailed breakdown of how it serves LLMs in-house rather than depending entirely on hosted APIs. The post covers model engine selection, API standardization across model versions, and deployment strategies including versioned rollouts and red-black deployments. The takeaway: at Netflix's scale, controlling the inference stack is worth the operational overhead, particularly for latency-sensitive and cost-sensitive use cases where per-token margins matter. For engineering teams evaluating whether to build or buy inference infrastructure, the post is a practical case study of what in-house serving actually requires in terms of engineering investment and ongoing maintenance.

Cloudflare released two new WAF rules in response to high-severity vulnerabilities reported by the WordPress security team. The rules protect all Cloudflare customers using affected WordPress versions, buying site operators time to apply upstream patches. WordPress powers a significant share of the web, and unpatched installations are a primary vector for defacements, malware distribution, and credential theft. If you run a WordPress site behind Cloudflare, the WAF rules are already active, but the real protection comes from updating WordPress itself. The vulnerabilities are serious enough that Cloudflare coordinated directly with WordPress on disclosure, which suggests they should be treated as urgent.

Mozilla.ai released transcribe.cpp, an open-source C/C++ speech-to-text inference library built on the ggml runtime. It supports multiple STT model families including Whisper and runs on Metal, Vulkan, and CUDA backends alongside a CPU path accelerated by tinyBLAS. The library makes local, private speech transcription feasible on consumer hardware without sending audio to a cloud API. For developers building voice interfaces, transcription features, or accessibility tools, this is a practical building block with no API keys, no per-minute pricing, and no data leaving the device. The project came out of Mozilla.ai's Builders in Residence program.

A major power outage in San Francisco on July 18 left Waymo's autonomous vehicles stranded in traffic, with multiple robotaxis blocking intersections after their batteries drained while inoperative. Waymo paused service for about an hour before resuming. San Francisco Mayor Daniel Lurie sent a letter to California regulators arguing the incident proves the current framework does not adequately address autonomous vehicle behavior during emergencies. The incident follows a pattern of Waymo vehicles becoming inoperable during power disruptions, and it raises a fundamental question for the autonomous vehicle industry: if robotaxis cannot handle a blackout without blocking city streets, what happens in a genuine disaster scenario?

Honda confirmed it will end sales of the Prologue, its only electric vehicle for the US market, after the 2026 model year. The Prologue was not a sales failure -- it was the sixth-best-selling EV in the US in 2025 with nearly 40,000 units sold. The discontinuation reflects a strategic shift toward hybrids and a broader industry retreat from all-electric commitments. The Prologue joins a growing list of canceled or killed-off EVs: multiple models from Ford, GM, and others have been axed or delayed as demand softens and policy uncertainty lingers. For buyers, it means fewer new EV options and raises practical questions about long-term parts and software support for discontinued electric models.