Visa Invests in Replit — Building the Infrastructure for AI Agents to Make Autonomous Payments
Visa has announced an investment in the developer platform Replit. The goal is to realize "agentic payments," in which AI agents carry out payments autonomously. Reportedly, more than 1,000 Visa employees already use Replit for prototyping and development.
A world in which AI agents buy products or pay for services on users' behalf is drawing closer. AI has so far been limited to processing information, but once it is connected to payments — the financial infrastructure — the practical utility of agents expands dramatically. The full-scale entry of Visa, the world's largest payment network, carries significant weight, and it's fair to say the foundations of the agent economy are falling into place.
Wix Lays Off 20% of Its Workforce — CEO Cites AI and Foreign Exchange
Wix, the website building platform, has announced a large-scale layoff affecting about 20% of its employees. The CEO has cited AI-driven efficiency gains and currency fluctuations as the main reasons.
According to TechCrunch, the CEO explained in an internal message that the introduction of AI tools has made it possible to handle the same workload with fewer people. As a case in which AI was explicitly named as a direct reason for job cuts, it is drawing attention across the tech industry.
The prevailing narrative has been that "AI doesn't take away jobs, it transforms them," but the Wix case demonstrates the reality that "AI reduces the number of people needed for the same work." It has also sparked discussion on Hacker News, and similar moves could spread to other companies.
Waymo's Chinese-Built Robotaxi "Ojai" Begins Commercial Service in California and Arizona
Waymo has begun commercial operations of its new robotaxi, the "Ojai," in California and Arizona. The vehicle is manufactured in partnership with a Chinese automaker and features a light blue body.
According to WIRED, the Ojai has entered full commercial operation, picking up general passengers. That Waymo, a subsidiary of Alphabet, has adopted Chinese-built vehicles is noteworthy amid continuing U.S.–China geopolitical tensions. In effect, Chinese EV makers' advantages in cost competitiveness and manufacturing capacity outweighed geopolitical considerations.
The autonomous taxi market is the scene of fierce competition, with not only Waymo but also China's Baidu and Pony.ai among others, and the deployment of the Ojai can be seen as an important step toward gauging how the market will evolve.
IBM and Red Hat Invest $5 Billion — Redefining the Future of Open Source AI with "Project Lightwell"
IBM and Red Hat have announced a $5 billion (approximately ¥750 billion) investment to redefine the future of open source in the AI era. The project is named "Project Lightwell."
Announced simultaneously through Red Hat's press release and IBM's newsroom, the initiative aims to prevent AI development from being monopolized by proprietary big tech. It seeks to build open source AI infrastructure, toolchains, and security foundations.
IBM has previously rolled out its own AI products such as Watsonx, but making a large-scale investment in the open source community in collaboration with Red Hat carries significant meaning from the perspective of democratizing AI development. It has also attracted attention, earning 4 points on Hacker News.
WhatsApp Releases "Incognito Chat with Meta AI" — Fully Private AI Conversations
WhatsApp has released the "Incognito Chat with Meta AI" feature. It is a mode in which conversations with Meta AI can be conducted completely privately, and chat content is reportedly never stored on servers or used for training.
Privacy concerns around AI assistants, stemming from worries about sharing personal information and confidential data with AI, have made many users hesitant to use them. WhatsApp's Incognito mode can be seen as an attempt to remove that barrier.
Meta is showing a proactive stance on privacy in AI chats and trying to differentiate itself from competitors such as ChatGPT and Google Gemini. The move appears to be a strategy of extending the privacy trust cultivated through end-to-end encryption into the AI space.
Zai Overhauls the Network for GLM-5.1 Inference — 33% Cost Reduction and 15% Throughput Improvement
Zai of China has deployed "ZCube," a new network architecture developed jointly with Tsinghua University and HarnetsAI, to its GLM-5.1 inference cluster (at a scale of 1,000 GPUs), achieving notable results.
By migrating from the conventional ROFT topology to a fully flat bipartite graph configuration, the company reduced switch and optical module costs by 33%, improved GPU inference throughput by 15%, and improved P99 first-token latency by 40.6%. What draws particular attention is that these results were obtained by changing only the network configuration — with the same GPUs, the same software stack, and the same model.
It solved the problem of asymmetric traffic caused by KV cache transfers in prefill-decode disaggregated inference by completely eliminating the spine layer. Performance improvements usually require hardware investment, but the fact that ZCube improved performance while actually lowering costs points to a new direction in the design of large-scale inference infrastructure.
---