El Agente Fármaco is a scalable AI scientist that turns plain-language drug-discovery goals into complex computational screening workflows by selecting and chaining tools. Notable for teams building agentic pipelines in pharma and scientific tool orchestration.

OpenAI is acquiring Pierre Computer Company to expand Codex's tooling for building, understanding, and reviewing code. Signals where OpenAI is investing in the coding-agent stack — worth a skim for anyone tracking competitive positioning in AI dev tools.

Proposes reference-free fault localization for RAG pipelines, addressing the hard problem of tracing errors across interdependent stages where faults propagate or hide behind correct answers. Practical for engineers debugging production RAG systems.

GitHub's official toolkit for spec-driven development (SDD) with AI coding agents. Backed by GitHub, so it may become a de facto standard for structuring agent-assisted engineering.

Presents a dual-track method for finding where MCP SDKs omit spec constraints, surfacing conformance gaps that could propagate into agent applications. Timely for teams building on Model Context Protocol, since SDK-level bugs are easy to miss until downstream.

Introduces LOLPlay, a benchmark of 36 champion-specific tasks in native League of Legends testing real-time visual agents via screenshots and mouse-keyboard control. Notable as a hard, dynamic environment for agentic reasoning-and-action coordination beyond static benchmarks.

A curated set of agent skills pulled straight from a well-known engineer's .agents directory. Worth a look as a reference for how practitioners actually structure reusable agent capabilities.

This issue covers Realtime-Bench findings that LLMs are very bad at low-latency tasks, plus Spacewar and Genie 4. Good roundup value for readers tracking agent capability gaps and the limits of realtime tasks.

Nathan Lambert reviews the state of AI benchmarking and why leaderboards like Artificial Analysis v5 struggle to capture what actually matters about model quality. Worth reading if you rely on benchmark rankings for model selection and want to understand their blind spots.

A security write-up exploring whether malicious content in log files can hijack an AI agent's behavior via prompt injection. Directly relevant to anyone deploying agents that ingest untrusted text, with a concrete attack surface to consider.