Lessons I learned while building my own coding agent from scratch.

Hacker News is a community-driven platform for sharing and discussing technology news, startups, and programming-related topics. Through user submissions and comments, Hacker News offers insights into emerging technology trends, industry developments, and entrepreneurial ventures. Readers can participate in discussions, share their insights, and stay informed about the latest advancements in technology and innovation.

Hacker News

Building a minimal coding agent from scratch revealed that simplicity beats complexity. A unified LLM API supporting multiple providers (Anthropic, OpenAI, Google, etc.) with cross-provider context handoff, combined with just four tools (read, write, edit, bash) and a sub-1000 token system prompt, performs comparably to feature-heavy alternatives on Terminal-Bench 2.0. The implementation includes a custom terminal UI framework using differential rendering and retained mode components. Key design decisions: no MCP support (use CLI tools instead), no built-in sub-agents (spawn via bash/tmux), no background processes (use tmux), and YOLO mode by default since security theater doesn't work when agents can execute code. Full observability and context control matter more than extensive tooling.

What I learned building an opinionated and minimal coding agent