Vol.32 · May 25, 2026
dera news AI Weekly Vol.32 | 2026-05-25 - This Week's AI News
🤖 dera news AI Weekly Vol.32
2026-05-25
This week's AI world in one sentence? AI is seeing rapid talent shifts and agent tech evolution, intensifying competition. Andrej Karpathy's move to Anthropic, new Google I/O announcements, and huge compute investments signal the full arrival of the 'agent era'.
📊 What You Need to Know This Week
This week, the most significant development in the AI industry is the move of Andrej Karpathy, OpenAI co-founder and former head of Tesla's AI division, to rival company Anthropic. This is an industry-shaking event that reveals what's happening at the forefront of AI research and development. His expertise is expected to significantly contribute to Anthropic's pretraining research using Claude, and his future contributions will be closely watched.
Google, at its annual I/O developer conference, unveiled the highly efficient AI model "Gemini 3.5 Flash" and the versatile "Omni." Flash, in particular, lowers the barrier to entry for AI agent tasks, opening new possibilities for business automation for small and medium-sized enterprises. Furthermore, the announcement of "Google Antigravity 2.0," a standalone desktop application poised to dramatically change AI development, suggests that the era of AI agent orchestration has truly begun.
The evolution of AI agents is also being driven by Chinese players. Alibaba's "Qwen3.7-Max" boasts 35 hours of continuous autonomous execution, signaling the arrival of an "agent era" where AI can plan, execute, and correct complex tasks over several days. Tencent has also open-sourced its AI agent memory system, "TencentDB Agent Memory," addressing issues of context loss and information recall to enable smarter AI applications. Microsoft, too, has released "Fara1.5," an AI agent for browser automation, demonstrating performance that surpasses existing major agents.
Meanwhile, practical challenges in AI adoption have also come to light. Anthropic's colossal $15 billion annual contract to use SpaceX's "Colossus" data centers underscores the immense computational resources required for large-scale AI development. Microsoft's decision to cut licenses for "Claude Code" for internal developers, encouraging a shift to its own "GitHub Copilot CLI," suggests that AI tool costs and strategic choices are major challenges for enterprises. Additionally, Elon Musk's lawsuit against OpenAI being dismissed and Pizza Hut franchisees suing Pizza Hut for over $100 million in damages due due to an AI delivery system highlight the legal and operational risks in AI implementation.
What we're watching closely is the rapid evolution of AI agent technology and the strategic corporate moves accompanying it. This evolution is not merely about providing tools; it holds the potential to transform business processes themselves, becoming a critical factor in establishing competitive advantage.
💡 This Week's Actions
1. Track key talent movements in the AI industry (10 min) The movements of top players in the AI industry are crucial indicators for predicting future technological trends and corporate strategies. The transfer of key figures like Andrej Karpathy provides clues to understanding a company's direction and focus areas. → OpenAI co-founder Andrej Karpathy announces he's joining Anthropic
2. Experiment with Google's latest AI agent development tools (30 min) Google's "Antigravity 2.0" and "Gemini 3.5 Flash" have the potential to accelerate business automation using AI agents. Antigravity 2.0, in particular, can link multiple AI agents to perform complex tasks, so consider trying out preview versions or related SDKs to see how they can be integrated into your business processes. → Google announces agent-optimized Gemini 3.5 Flash and a do-anything model called Omni → Google Launches Antigravity 2.0 at I/O 2026
3. Reconfirm AI agent security and integration (20 min) Anthropic's self-hosted sandboxes and MCP tunnels for Claude Managed Agents represent significant progress in securely integrating AI agents with internal corporate APIs. Companies considering or already implementing AI should evaluate the latest security measures to achieve safe integration while minimizing the risk of credential leakage. → Claude agents can finally connect to enterprise APIs without leaking credentials
🧩 Shikumika Service | Organizing People-Dependent Workflows
If you have business processes that "only Ms. Tanaka understands," dera's Shikumika service might help. One person handles interviews → design → build → adoption, end-to-end. Starting at ¥100K/month, first-month refund guarantee.
👉 Book a 15-min free consultation
📰 This Week's AI Articles (11 stories)
1️⃣ OpenAI co-founder Andrej Karpathy announces he's joining Anthropic
🏷️ Talent Movement What happened? Andrej Karpathy, an OpenAI co-founder and former head of Tesla's AI division, announced he is joining rival company Anthropic. He will return to R&D at Anthropic, specifically building a team focused on accelerating pretraining research at the frontier of large language models (LLMs). Our take The flow of top talent in the AI industry significantly impacts company strategies and technological trends. It will be interesting to see how Karpathy's expertise contributes to Anthropic's Claude development and how this move alters the competition between OpenAI and Anthropic. 📎 Read more
2️⃣ Here's why Elon Musk lost his suit against OpenAI
🏷️ Lawsuit What happened? In the lawsuit brought by Elon Musk against OpenAI, the court dismissed Musk's claims. The dismissal occurred before the merits of the case could be heard, due to the suit being filed too late. Musk had alleged that OpenAI broke its non-profit promise by establishing a for-profit subsidiary. Our take This lawsuit brings a close to a major AI industry debate surrounding OpenAI's founding principles and its subsequent transition to a for-profit entity. The outcome serves as an important case study for considering the governance and mission of AI companies. 📎 Read more
3️⃣ Google announces agent-optimized Gemini 3.5 Flash and a do-anything model called Omni
🏷️ Model Release What happened? Google announced its latest AI models, "Gemini 3.5 Flash" and "Omni." Flash, in particular, offers cutting-edge intelligence with high efficiency, potentially making large-scale AI agent tasks more cost-effective to implement. Our take Google's AI models have evolved significantly in just one year. The efficiency of Gemini 3.5 Flash, especially, will greatly lower the barrier for small and medium-sized businesses to integrate AI agents into their operations. It is expected to be utilized for automating a wide range of tasks, including customer support and data analysis. 📎 Read more
4️⃣ Google Launches Antigravity 2.0 at I/O 2026
🏷️ Development Platform What happened? At its I/O 2026 developer conference, Google unveiled "Google Antigravity 2.0," a new standalone desktop application designed to fundamentally change AI agent-driven development. It focuses on "agent orchestration," allowing multiple AI agents to collaborate on tasks. Our take Antigravity 2.0 is a tool that elevates AI agent-based business automation to a new level. With scheduling features and integration with other services, it has the potential to directly boost productivity for small and medium-sized businesses. This move to democratize AI agent development is noteworthy. 📎 Read more
5️⃣ Anthropic is paying $15 billion a year for access to Elon Musk’s data centers
🏷️ Infrastructure Investment What happened? AI development firm Anthropic has reportedly signed a deal to use Elon Musk's SpaceX AI data centers "Colossus" for $15 billion annually. This massive contract amounts to $1.25 billion per month through May 2029. Our take This enormous contract reiterates the vast computational resources required for large-scale AI model development. High-performance data centers are essential for AI technology advancement, indicating that the development race among major AI companies extends to infrastructure investment. 📎 Read more
6️⃣ Alibaba's Qwen3.7-Max achieves 35 hours of autonomous AI execution
🏷️ Agent Technology What happened? Alibaba's Qwen team announced that its AI model "Qwen3.7-Max" achieved approximately 35 hours of continuous autonomous execution. This symbolizes the arrival of an "agent era" where AI can autonomously plan, execute, and correct complex tasks over several days, going beyond mere text generation. Our take The era of AI autonomously handling complex tasks for days is approaching. Small and medium-sized business leaders should adopt a strategic perspective on how to integrate AI agents into their future business processes and link them with existing systems. 📎 Read more
7️⃣ Tencent Open-Sources TencentDB Agent Memory: A 4-Tier Local Memory Pipeline for AI Agents
🏷️ Agent Infrastructure What happened? Tencent has open-sourced "TencentDB Agent Memory," a memory system for AI agents. This system combines short-term memory with a hierarchical long-term memory to solve problems like "context loss" and "recall failure," where AI forgets past conversations or crucial information. Our take This system, which enhances AI agent memory and efficiency, is highly beneficial for companies operating AI systems that require long-term dialogue or continuous information reference, such as customer support or sales assistance. It is noteworthy as an improvement that directly leads to reduced operational costs. 📎 Read more
8️⃣ Microsoft Releases Fara1.5: A Family of Browser Computer-Use Agents
🏷️ Agent Technology What happened? Microsoft Research announced "Fara1.5," a family of browser computer-use AI agents. This agent demonstrates performance that significantly surpasses existing major agents like OpenAI Operator and Google Gemini 2.5 Computer Use. Our take High-performance browser automation AI agents like Fara1.5 have the potential to automate various tasks for small and medium-sized businesses, including data entry on websites, information gathering, and routine operations. When these technologies become available, the AI transformation of routine work will accelerate. 📎 Read more
9️⃣ Claude agents can finally connect to enterprise APIs without leaking credentials
🏷️ Enterprise AI What happened? Anthropic announced new capabilities for Claude Managed Agents: "self-hosted sandboxes" and "MCP tunnels." These allow AI agents to securely connect to internal corporate APIs without the risk of credential leakage, accelerating AI adoption. Our take A major security challenge in enterprise AI adoption is being addressed by Anthropic's new features. By reducing the risk of information leakage while promoting AI-driven business automation, a path for small and medium-sized businesses to safely deploy AI agents has opened. 📎 Read more
🔟 Microsoft Cuts Claude Code Licenses Internally
🏷️ Cost Management What happened? Microsoft is reportedly significantly reducing licenses for Anthropic's AI coding tool "Claude Code," which was provided to internal developers, and encouraging a transition to its self-developed "GitHub Copilot CLI." It is suggested that internal usage may have exceeded budget. Our take The move by a major tech company to consolidate around its own AI tools shows that even excellent AI technologies from other companies are chosen based on a company's own ecosystem and cost strategy. When selecting AI tools, factors beyond functionality, such as compatibility with the ecosystem and cost, become crucial. 📎 Read more
11️⃣ Pizza Hut Franchisee Sues Over AI Delivery System, Claiming $100M+ Damages
🏷️ AI Adoption Risk What happened? Pizza Hut franchisees have sued Pizza Hut, claiming over $100 million in damages due to an AI delivery system mandated by the corporate office. The AI system allegedly negatively impacted driver behavior, leading to delivery delays and decreased customer satisfaction. Our take This case illustrates that AI adoption does not always succeed and can have unexpected pitfalls. It re-emphasizes the importance of thoroughly evaluating the impact on operations, training, and potential risks, and proceeding cautiously when implementing AI systems. 📎 Read more
📚 Editor's Note
This week was packed with news that truly symbolizes the dynamism of the AI industry. Andrej Karpathy's move to Anthropic highlights what top talent is pursuing at the frontier of AI, and the announcements at Google I/O make us realize that AI agents are beginning to deeply integrate into our work and lives.
On the other hand, Anthropic's massive compute resource contract, Microsoft's strategic choice of internal AI tools, and the Pizza Hut lawsuit surrounding an AI system confront us with the real-world challenges, costs, and risks of AI implementation and operation. AI is not a magic wand; it requires appropriate strategy, careful implementation, and continuous evaluation.
We'll be back next week with more useful insights and thought-provoking content. dera news editorial team
🤝 Ready for Team-Wide AI Adoption?
Once you've tested ChatGPT individually and felt its potential, the next step is team-wide implementation. Let's talk about your specific challenges and goals. 📩 Get in Touch
📬 About this newsletter