本文由 AI 分析生成
建立時間： 2026-03-28 來源： https://simonwillison.net/2025/Oct/7/vibe-engineering/#atom-everything
Summary
Simon Willison’s October 2025 post proposing “vibe engineering” as a term for the responsible, expert-led end of the AI coding spectrum — contrasted with “vibe coding” (fast, irresponsible, no accountability). Lists 12 existing software engineering practices that LLMs actively amplify. Updated Feb 2026: the term “Agentic Engineering” appears to be winning out.
Simon Willison 2025 年 10 月提出「vibe engineering」作為 AI 輔助程式設計負責任端的術語——與「vibe coding」（快速、不負責任）形成對比。列出 12 種 LLM 能放大效益的現有軟體工程實踐。2026 年 2 月更新：「Agentic Engineering」一詞似乎正在勝出。
Key Points
- Vibe coding: fast, prompt-driven, irresponsible AI coding — outsource task to LLM, accept result if it appears to work
- Vibe engineering (proposed): senior professionals accelerating work with LLMs while staying “proudly and confidently accountable” for the software they produce; now called “Agentic Engineering” (Feb 2026)
- Parallel agents: running multiple coding agents simultaneously on different problems — already doing this himself; “surprisingly effective, if mentally exhausting”
- 12 practices LLMs reward (selected):
- Automated testing — agents can fly with a robust test suite; without tests, agent may claim success without actual verification
- Planning in advance — iterate on plan first, hand off code generation to agent
- Comprehensive documentation — feeds context to LLM; good docs enable building matching implementation from input alone
- Version control and small commits — essential for reviewing and reverting agent changes
- Code review culture — necessary to catch agentic errors
- Security awareness — LLMs can miss security implications
- Modular architecture — LLMs work better with well-decomposed systems
Insights
The insight that “LLMs actively reward existing top-tier software engineering practices” reframes the AI coding debate. Rather than AI lowering the bar, it raises the ROI on practices good engineers already know but teams often skip under time pressure. The terminology battle (vibe engineering → agentic engineering) is itself informative: the field is groping for language to describe a new mode of professional work. Willison’s acknowledgment that parallel agent work is “mentally exhausting” is an honest counterpoint to the productivity hype.
Connections
Raw Excerpt
One of the lesser spoken truths of working productively with LLMs as a software engineer on non-toy-projects is that it’s difficult. There’s a lot of depth to understanding how to use the tools, there are plenty of traps to avoid, and the pace at which they can churn out working code raises the bar for what the human participant can and should be contributing.