SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsthree-stage-evolution

three-stage-evolution

General & other

View repository

Updated 10-15 23:44

About this skill

Source Metadata

Original cangjie-skill frontmatter from the distillation run:

name: three-stage-evolution
description: |
  评估个人或团队在 AI 工具使用上的进化阶段,给出下一步提升建议。
  当用户想判断"我现在 AI 用得怎么样"、"下一步该学什么"、或团队要做 AI 能力评估时使用。
  不适用于: 已经处于 Loop 阶段需要优化具体系统的场景。
  关键 trigger: "我现在在哪个阶段"、"AI 使用下一步"、"团队 AI 能力评估"。
source_book: "Loop Engineering 视频合集"
source_chapter: 视频3 (小木头) / 视频2 (Boris Cherny)
tags: [evolution, self-assessment, capability-model, growth]
related_skills: [loop-build-path, loop-worthiness-test, comprehension-gap]
Three-Stage Evolution Model — Locate Your AI Usage Stage

R — Reading (Original Text)

"First stage, line-by-line driving... you hand-write code and the model auto-completes for you. Second stage, parallel manual, you might run 5 or 10 conversations at the same time, each one working. But every conversation is initiated by you personally... In the third stage you no longer initiate conversations; you wrote a system that lets it read your repository on its own, read issues, read CI failures, and decide on its own what kind of prompts to use to drive agents to work." — Xiaomutou (Video 3)

I — Interpretation (Own Words)

The way individuals collaborate with AI goes through three evolution stages:

  1. Line-by-line driving (Prompting): Hand-write code/content, AI provides auto-completion. The person is the operator, AI is the tool.
  2. Parallel manual (Parallel): Run 5-10 conversations at the same time, each one working. The person is the dispatcher, switching between multiple windows.
  3. Loop system (Loop Engineering): No longer initiate conversations; design a system that lets AI work autonomously. The person is the system designer.

Characteristics of each stage:

  • Stage 1: Every task is one-off, AI is "advanced auto-completion"
  • Stage 2: Efficiency improves but the person is the bottleneck; all conversations must be initiated by hand
  • Stage 3: The system runs autonomously; the person only needs to design and supervise

A1 — Past Application (Cases in the Book)

Case 1: Boris Cherny (Video 2)

  • Described the evolution from "line-by-line driving" to "writing loops"
  • Currently at Stage 3: has a bunch of loops running, and only writes new loops himself

Case 2: Xiaomutou (Video 3)

  • Self-assessed as being at Stage 2 (parallel manual)
  • Transitioning toward Stage 3 (demonstrated a topic-selection loop)

Case 3: Adam Gillock (Video 1)

  • Non-technical background, but already uses loops for video editing
  • Shows that Stage 3 is not limited to technical people

A2 — Future Trigger

  1. When self-assessing: "How well am I using AI right now?"
  2. When making a learning plan: "What should I learn next?"
  3. When building team capability: Assess the team's overall AI usage stage
  4. When explaining Loop Engineering to others: Use this model to explain "why upgrade to Stage 3"

Language signals: "what stage am I at right now", "next step for AI use", "team AI capability assessment", "from manual to automatic"

Differences from adjacent skills:

  • loop-build-path: Construction guide after upgrading to Stage 3 (this skill is positioning and decision-making)
  • loop-worthiness-test: In Stage 3, determining whether a specific task should be turned into a loop (this skill is overall stage assessment)
  • comprehension-gap: Risks of Stage 3 (this skill is the advancement path of Stage 3)

E — Execution (Executable Steps)

Step 1: Self-assess the current stage

Answer the following questions:

  • Do you personally initiate a conversation every time you use AI? → Stage 1 or 2
  • Do you have multiple AI conversations running at the same time? → Stage 2
  • Do you have scheduled/event-triggered AI tasks? → Possibly Stage 3
  • Have you designed a system that "lets AI decide autonomously what to do"? → Stage 3

Step 2: Locate by matching characteristics

  • If it is mainly "human initiates → AI executes → human gives feedback" → Stage 1
  • If it is mainly "a human manages multiple AI tasks at the same time" → Stage 2
  • If it is mainly "the system runs automatically, and the human designs the system" → Stage 3

Step 3: Give next-step recommendations

  • Stage 1 → Stage 2: Learn to use multiple AI tasks in parallel, isolating them with worktrees
  • Stage 2 → Stage 3: Choose a small task and use loop-build-path to make the first loop
  • Stage 3 → Optimization: Use loop-5plus1-architecture to audit the existing system

B — Boundary

Scenarios in which this skill should not be used:

  1. Already at Stage 3 and running well: What is needed is optimization, not stage assessment
  2. Purely technical questions: "Why is this loop throwing an error?" — this is not stage positioning
  3. Non-AI scenarios: This model only applies to AI tool usage

The author's blind spots and limitations:

  • The three-stage model is a linear assumption; in actual use, multiple stages may be mixed
  • "Stage 3 is best" is an implicit assumption, but the author also admits that "most tasks do not need a loop"
  • The video's authors are all at Stage 3 or are upgrading, lacking the perspective that "Stage 1 is also fine"

Adjacent but easily confused methodologies:

  • Dreyfus model: five stages from "novice → expert"; this skill is three stages of AI usage
  • Technology adoption life cycle: "innovators → early adopters → ..."; this skill is a personal capability model