What I learned building an opinionated and minimal coding agent

December 15, 2025

Table displaying performance metrics for the agent "pi (claude-opus-4-5)" on the "terminal-bench" dataset, including 428 trials, 71 errors, a mean score of 0.479, reward distribution with 213 successes and 215 failures, and a breakdown of exception types and counts.

I’ve also built a bunch of agents over the years, of various complexity. For example, Sitegeist, my little browser-use agent, is essentially a coding agent that lives inside the browser. In all that work, I learned that context engineering is paramount. Exactly controlling what goes into the model’s context yields better outputs, especially when it’s writing code. Existing harnesses make this extremely hard or impossible by injecting stuff behind your back that isn’t even surfaced in the UI.

Source: What I learned building an opinionated and minimal coding agent

Mario Zechner built his own minimal coding agent. Think of a lightweight version of Claude Code or OpenAI’s Codex. You can follow along here.