Articles
Technical insights on AI agents, tool patterns, and practical implementations
Running Qwen 3.8 Flash Next Locally: Strix Halo, Strata, and Gufo
Two weeks of testing Qwen 3.8 Flash Next on a Strix Halo mini-PC and an NVIDIA workstation, across three inference engines - llama.cpp, Strata, and Gufo. The 42% tool-call failure, the fix, and why we test and why evals are important.
GuideAnts Quick Setup and Tour
A complete, screen-by-screen walkthrough of installing GuideAnts - the free, open AI operating system - with Docker as the only prerequisite, configuring OpenRouter, and touring the out-of-the-box features. The whole demo cost ten cents.
Is MCP Really Necessary? Why "APIs Are Hard" Is Overstated and MCP Is Oversold
Why the common 'APIs are hard' pitch for MCP is overstated, what you can do with direct APIs + LLMs, and the real tradeoffs MCP introduces for cost and quality.
How to Lower Cost and Improve Quality with Tool-Based Agents
A deep dive into the Search + ReadWeb design pattern used in GuideAnts Notebooks, showing how tool-based agents reduce tokens, improve answer quality, and differ from typical MCP setups.
Faster, Cheaper, Better: Why Guides and Assistants Beat Single-Model AI
Learn how guides and assistants create modular AI workflows that deliver quality, speed, and cost efficiency. Real-world example: creating a podcast video with Python using multiple specialized assistants.