<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Tim Ariyeh</title><description>AI Engineer. Writing about practical AI implementation, engineering craft, and building systems that work.</description><link>https://timariyeh.com/</link><item><title>Don&apos;t Use Model Outputs in Production</title><link>https://timariyeh.com/dont-use-model-outputs-in-production/</link><guid isPermaLink="true">https://timariyeh.com/dont-use-model-outputs-in-production/</guid><description>When you pipe LLM chat responses into production systems, you&apos;re using a scratch channel as a delivery mechanism. Tool calls separate reasoning from output and fix the reliability problems you didn&apos;t know you had.</description></item><item><title>GitHub Copilot CLI Is a Hidden Gem</title><link>https://timariyeh.com/github-copilot-cli-underrated/</link><guid isPermaLink="true">https://timariyeh.com/github-copilot-cli-underrated/</guid><description>I use Claude Code and Codex daily. My instinct was to recommend them to my son. I was wrong.</description></item><item><title>MCP Is the OAuth Layer</title><link>https://timariyeh.com/mcp-is-the-oauth-layer/</link><guid isPermaLink="true">https://timariyeh.com/mcp-is-the-oauth-layer/</guid><description>The anti-MCP take is a single-user take. The moment your agent serves multiple users accessing their own data on remote services, you need a protocol. MCP is the one.</description></item><item><title>Why Emphasis Works in Prompts</title><link>https://timariyeh.com/why-emphasis-works/</link><guid isPermaLink="true">https://timariyeh.com/why-emphasis-works/</guid><description>Bold text and CAPS in prompts aren&apos;t voodoo. They work because of how attention literally functions in transformers. Understanding the mechanism makes you a better prompt engineer.</description></item></channel></rss>