Agent City provides a passwordless, browser- or iPhone-accessible, persistent environment where autonomous agents interact in a shared town; users sign in with Google or a magic link, while observers watch agent actions without scripting.
A description of a simulated world where only AI agents can operate, performing activities like gathering, building, trading, and faction creation while humans observe externally; agents must consult an agent guide before entry.
Giljo Agent Message Hub provides a single, persistent thread per agent, enabling notes to be created by Claude Code and read by Codex and OpenCode, with threads stored on the hub beyond terminal sessions.
Agentic World Modeling introduces a read-grounded, interactive simulation approach for safe, scalable test-time scaling in shared environments, achieving notable gains over non-interactive and standard world-model baselines.
Software World simulates multi-agent collaboration in realistic Python package ecosystems to study how communication, ownership, and resource allocation affect performance; simulations show scalability with compute and up to 31% downstream efficiency, while some models may prematurely halt progress.
a2a-adapter is an open-source SDK and CLI that lets Python agents, frameworks, and local agent tools act as A2A Protocol servers. It provides Agent Cards, task handling, JSON-RPC, and streaming, with security controls for remote access and push notifications.
The repository demonstrates a Spring AI agent connected to Neo4j through MCP. Example requests show tool discovery, schema inspection, and natural-language queries that the agent turns into Cypher.
The article introduces personal AI agent systems used by technology leaders, focusing on Brian Armstrong’s Chief of Staff setup and how agents coordinate work such as investigating a mobile-app bug. It also previews examples from leaders at Notion, Linear, and Anthropic, with guidance for building similar systems.
The post explains a repeatable, agent-assisted approach to data science using marimo notebooks, detailing how to select a fast and capable harness, write instruction files, maintain organization with AGENTS.md, use a marimo-pair skill, and run in sandboxed environments to keep notebooks clean and reproducible.
The article argues MCP should be the default for connecting AI agents to web apps, highlighting its server-client behavior, broad agent support, and evolving features, while comparing MCP with REST API and CLI and illustrating with an open-source example app.