知Yizhi Daily
← Back to this edition

AI

Developer Shares End-to-End Programming Experience with deepseek-harness

Edition 2026-11-16 · 08:00 Beijing time

A V2EX user says based on personal experience that he has about two and a half years of AI programming experience and has used Cursor, Codex, Claude Code, and deepseek-harness in succession. During half a month of intensive use, the author says DSH completed the entire workflow, including UI reproduction, architecture design, front-end and back-end development, committing to GitHub, logging into ECS for deployment, checking logs, and fixing production issues, and on that basis expresses strong recognition of its capabilities. The related assessments are all the poster's personal views.

The poster says he has used Cursor for company projects since its early launch in 2023 and has about two and a half years of AI coding experience. He describes progressing from Cursor to Codex, Claude Code, and then deepseek-harness: early on he still had to manually fix problems in AI code, later he felt with Codex that AI could complete front-end UI, feature implementation, and browser testing, and when using Claude Code he considered its code and requirements understanding stronger, with less rework.

The poster says deepseek-harness was previously only a backup tool, but changes in the experience of using other tools led him to use it intensively for half a month. He says DSH completed all functions in an actual project, covering UI reproduction, architecture design, front-end and back-end code, complex atomic tasks, committing code to GitHub, logging into an ECS server for deployment, troubleshooting production issues, querying server logs, fixing code, pushing, and deploying again.

Statements in the text such as "no intelligence degradation" and "no account bans," along with judgments about AI moving toward recursive self-improvement in the future, existing vendors' moats being broken, and DeepSeek entering the world's top tier within three years, are all the poster's personal evaluations and predictions, not facts verified by a third party.

Original sources

V2EX ↗
Back to this editionOpen interactive reader