Anthropic Acquires Stainless — Expanding Agent Reach Through SDKs and MCP

On May 16, Anthropic announced the acquisition of Stainless, the leading company in SDK and MCP server tooling. Founded in 2022, Stainless has technology that auto-generates SDKs in multiple languages — TypeScript, Python, Go, Java, Kotlin, and more — from API specifications, and it has consistently generated Anthropic's official SDKs.

Katelyn Lesse, Anthropic's Head of Platform Engineering, said: "Agents are only as useful as what they can connect to. By bringing on the Stainless team, we can further enhance Claude's ability to connect to data and tools."

The AI frontier is now shifting from "models that respond" to "agents that act." SDKs and MCP (Model Context Protocol) are the foundational technologies of that shift, and the acquisition can be read as Anthropic taking a key piece of its agent strategy in-house. Given that hundreds of companies depend on Stainless-generated SDKs, the impact could extend well beyond Anthropic.

Pope Leo XIV to Publish First-Ever Encyclical on AI, Anthropic Co-founder Invited

It has been announced that Pope Leo XIV will publish his first encyclical on AI on May 25. An encyclical is a pope's most authoritative document in the Catholic Church, and this will be the first in history to make AI its subject.

What draws particular attention is that Anthropic co-founder Christopher Olah has been invited as a guest speaker. It is highly unusual for an AI company founder to be invited to an official papal event, a sign of how strongly the religious world is focused on AI's social and ethical impact.

Since taking office, Pope Leo XIV has repeatedly spoken about AI's impact on human dignity, and the encyclical is expected to lay out ethical guidelines for AI development.

Anduril and Meta Are Building Military AR Smart Glasses — Directing Drone Strikes with a Gaze

Defense tech company Anduril has unveiled details of a military AR headset developed jointly with Meta. According to MIT Technology Review, the headset is designed to let soldiers direct drone strikes using gaze tracking and voice commands.

Anduril VP Quay Barnett (formerly of Army Special Operations Command) says the fundamental goal is "optimizing the human as a weapons system," envisioning drones and soldiers sharing visual information and making decisions as an integrated unit.

Two projects are currently underway. The first is the Army's SBMC (Soldier Born Mission Command) program, for which Anduril has already secured a $159 million prototyping contract. The second is the self-funded development of its own helmet and headset, "EagleEye." As military applications of AI accelerate, the need for ethical debate is growing too.

Qwen 3.7 Release Is Imminent — Anticipation Builds in the LocalLLaMA Community

The release of "3.7," the latest version of Alibaba's Qwen series, is just around the corner. In the LocalLLaMA community, expectations are running high, with reports that 3.7 has dropped in Qwen Chat and the development team posting that they "can't wait for the 3.7 release."

Its predecessor, Qwen 3.5 (35B-A3B), has earned praise for its agentic coding performance in local environments. A user who ran it via llama.cpp on an RTX 4090 + 5060 Ti setup as a backend for Claude Code rated it "better than Gemma 4 26B." Observers note it is particularly interesting that it shines in agentic coding rather than in chat UIs.

Some users are also running Qwen 3.6 on a Tesla P40. Hardware compatibility challenges are being discussed as well, such as struggling with KV cache quantization to enable MTP (Multi-Token Prediction) inference.

Dropbox Open-Sources "Witchcraft," a Fast SQLite-Based Semantic Search Engine

Dropbox engineers have open-sourced "Witchcraft," a from-scratch reimplementation of Stanford's XTR-Warp semantic search engine written in Safe Rust.

Its biggest feature is that it uses a single SQLite file as its backend and runs entirely locally — no API keys, no vector database, no chunking required. On an Apple MacBook Pro M2 Max, it achieved 20ms p.95 latency on the NFCorpus benchmark with an NDCG@10 of 33%, delivering accuracy on par with the original XTR-Warp (running on server-class hardware) at more than twice the speed.

The project also includes a CLI tool called "Pickbrain" that indexes Claude Code and OpenAI Codex session histories and memory files for fast semantic search. For individual developers, the local AI infrastructure looks set to get even richer.

Devin AI Announces Automatic Issue Triage Feature

AI coding agent Devin has announced an automatic triage feature for GitHub issues. The new "Auto Triage" feature has Devin automatically analyze and classify issues posted to a repository, helping with prioritization and assignment.

Coding agents have mainly starred in "writing code" so far, but issue triage marks a full-fledged entry into maintenance and operations. Classifying issues alone consumes significant engineering effort in large repositories, so demand for AI-driven automation here is high.

---