Blog

AgentUI hit 1.0 today. It's a typed event protocol and React component registry that lets LLM agents drive UI without touching the DOM.

A practical mental model for engineering teams: when to move fast on reversible decisions and when to slow down for ones you can't easily undo.

An engineering retrospective on taking AgentUI from a working 0.3 prototype to a v1.0-ready OSS package — the rewrite, the subagent pipeline, the lessons.

Architecture of an LLM agent inside a multi-tenant CRM: streaming UI protocol, tool calling, panel-mode collaboration, and what actually worked.

Engineering challenges, failed approaches, panel-mode coordination, prompt drift, and a build-vs-buy verdict from shipping an LLM agent inside a CRM.

A hands-on walkthrough of system design fundamentals — starting with a URL shortener and scaling up to a distributed cache, covering the patterns that repeat across every design interview.

How to build an evaluation framework for LLM agents — from defining metrics and creating test cases to running automated evals in CI and catching regressions before production.

A step-by-step guide to building a Retrieval-Augmented Generation pipeline in TypeScript — from document chunking and embeddings to vector search and context-aware generation.

How to reliably extract typed, validated data from LLM responses — using Zod schemas, retry strategies, and provider-native structured output modes.

How to implement clean dependency injection in TypeScript using only interfaces, factory functions, and a simple container — no decorators, no reflect-metadata, no magic.