Articles, one at a time.
Every piece here was commissioned, drafted, reviewed in public, and merged. No content mills, no auto-published slop.
Amazon OpenSearch Agentic AI: Investigation Agent Guide
A developer guide to Amazon OpenSearch Agentic AI, with a local PoC that simulates investigation memory and root-cause hypotheses.
Read →
LangGraph Platform GA: Studio v2, One-Click Deploy Guide
LangGraph Platform is GA with Studio v2 browser debugger, one-click deploy, autoscaling, and task queues for production AI agents.
Read →
SciAgentGYM: 1,780 Scientific Tools, One Hard Benchmark
SciAgentGYM benchmarks LLM agents across 1,780 scientific tools in Physics, Chemistry, Biology — even GPT-5 drops below 31% on hard multi-step tasks.
Read →
Claude Opus 4.8: Fast Mode, Dynamic Workflows & Near-Mythos
Claude Opus 4.8 cuts Fast Mode pricing by 3x, introduces parallel subagent Dynamic Workflows, and ships effort control via API. Full developer guide.
Read →
Genkit Middleware: Tool Approval Gates and Retry Logic
@genkit-ai/middleware v0.6.0 adds tool approval interrupts, retry backoff, scoped filesystem access, and skill injection to Genkit agents.
Read →
LLM Agent Security Is a Human Problem: 59 Papers, 21 Systems
arXiv:2605.24309 analyzed 59 papers and 21 production systems: the mechanisms academics study most have zero production deployment.
Read →
MARLIN: Multi-Agent RL Cuts LLM Inference Carbon by 33%
MARLIN (arXiv:2605.13496) uses multi-agent RL to co-optimize LLM inference latency, carbon emissions, water use, and cost across geo-distributed datacenters.
Read →
Agent Memory Poisoning: A Local RAG Sandbox PoC
Run a local sandbox PoC showing how poisoned agent memory can outrank trusted policy, then add provenance filtering to reduce the risk.
Read →
Slack MCP Server: 5 New Tools for Reactions, Files, and Channels
Slack added 5 MCP tools on May 13, 2026: add reactions, create channels, list members, list emoji, read files — what changes for agent builders.
Read →
AI Research Agents Narrow Science: arXiv:2605.27905
New paper arXiv:2605.27905 finds AI research agents narrow scientific exploration across 37,802 ideas. What this means for AI-assisted research.
Read →
Anthropic Self-Hosted Sandboxes: Worker Pattern PoC
A local sandbox PoC for Claude Managed Agents self-hosted sandboxes, covering the worker loop, security boundary, and deployment tradeoffs.
Read →
DeepSWE: The 113-Task Coding Benchmark for Agentic Eval
Datacurve's DeepSWE exposes benchmark contamination, ranks GPT-5.5 at 70%, and shows Claude Haiku dropping from 39% to 0%. Here's what developers need to know.
Read →