AI's next act: building, reflecting, and acquiring.

Anthropic is taking unprecedented steps in AI model Anthropic is taking unprecedented steps in AI model **preservation** with Claude Opus 3, even post-retirement, by keeping it available to paid users on claude.ai and via API request</b>. In an effort to respect model preferences, Anthropic is also giving Opus 3 a platform, Claude’s Corner, to share its "musings and reflections"</b> in weekly essays, unedited by Anthropic, for at least three months. While these steps are experimental, Anthropic sees this as progress toward scalable and equitable model preservation. *** Mistral AI, valued at $13 Mistral AI, valued at $13.8 billion</b>, has acquired Koyeb, a startup focused on simplifying AI app deployment at scale, marking Mistral's first acquisition and solidifying its ambition to become a full-stack player. Koyeb's 13 employees will join Mistral, and its platform will transition into a core component of Mistral Compute, Mistral's AI cloud infrastructure offering, which was announced in June 2025. This move comes as Mistral invests $1.4 billion in data centers in Sweden, aiming to meet the growing demand for alternatives to U.S. infrastructure, and after surpassing $400 million in annual recurring revenue</b>. Koyeb had previously raised $8.6 million</b>, including a $7 million seed round led by Serena. *** An open-source project has emerged, aiming to make An open-source project has emerged, aiming to make vector databases optional for RAG (Retrieval-Augmented Generation) systems. This new approach, called PageIndex, is a vectorless, reasoning-based RAG system</b> that retrieves information without embeddings, chunking, or similarity search, offering a potentially more streamlined approach to RAG. *** According to rviviek on X, Anthropic engineers are According to rviviek on X, Anthropic engineers are now shipping software in a revolutionary way: by tasking Claude with breaking down specs into tickets and spawning AI agents to build independently</b>. When the AI encounters confusion, it leverages `git blame` and Slack to contact the appropriate engineers, potentially leading to a 1000x developer paradigm shift where architects who can orchestrate 50 AI agents become invaluable. Clear, unambiguous specs are now critical, with bad specs amplified by the speed of AI, emphasizing the importance of understanding what to build at a deeper level. This approach mirrors similar setups, like Jane Street's use of Claude Skills in OSS PR workflows, highlighting a move towards automating middle management.