Blog
Return Label OCR: Which Large Language Model Comes Out on Top?
Slow return-label recognition in GWMS prompted a comparison of vision models using the same prompt and API approach. Here are the original reports from DeepSeek Harness and Claude Code CLI, and why the difference matters.
I Had DeepSeek Harness and Claude Code CLI Do the Same Thing, and Let Codex Judge
A head-to-head test of DeepSeek Harness and Claude Code CLI on the same task — same starting point, converging results — with Codex reviewing both sides' code changes as the judge.
I Asked DeepSeek Harness to Give Me an Interview Question
DeepSeek Harness turned a GitHub job posting into a 'Bounded Blocking Queue' concurrency interview question; I hand-wrote an attempt (bugs included) and had AI review it — a real usage log.
DeepSeek Harness Isn't a Claude Code Clone — It's a Strike From Another Dimension
Starting from each vendor's official definition and the Kubernetes analogy: DeepSeek Harness is an agent harness (foundation), while Claude Code and Codex are coding agents (products) — a strike from a different dimension.
How this site is built: Astro 7 bilingual blog + Cloudflare Pages
From domain to launch: stack choices, the bilingual i18n structure, content collections, deployment flow, and how to publish new posts.
Welcome: why HarnessDo.com exists
Why this blog exists: tracking every step of DeepSeek Harness development, and documenting my own learning and growth. Plus the Cloudflare domain & hosting story.