Research notebook · started 2026-09-30
MDLC Industry Research
Research on the evolving model development lifecycle across ML, GenAI, and agentic systems.
How this repository works
Each page sits at one layer of the research progression. Claims move rightward only with citations. Every statement is labelled as a fact, observation, interpretation, hypothesis, open question or decision.
How to use this notebook · 每天怎么用
| When · 频率 | What to do · 做什么 | Say to Claude · 对 Claude 说 |
|---|---|---|
| Daily · 每天 | Ask one question. It is sharpened and logged as the next Q. 问一个问题，磨尖后记为下一个 Q。 | “我想到一个新问题：…” |
| When promoted · 值得深挖时 | Research it into a cited, bilingual note — only if it meets 2 of 3: evidence is findable · it affects a hypothesis (H) · you care about the answer. 满足 3 条中的 2 条才写成笔记：能查到证据 · 影响某个假设 · 你关心答案。 | “这个问题值得研究吗？” / “开始研究 Q5” |
| Every note · 每篇笔记 | Check whether it supports or contradicts an existing H; note it, don't rewrite the thesis. 只标注它支持或反驳哪个 H，不改论点页。 | (automatic · 自动) |
| Monthly · 每月 | Review the thesis: collect evidence per hypothesis, adjust confidence, record the revision. 复查论点：汇总每个假设的证据，调整信心，记录修订。 | “帮我复查一下论点” |
Current thesis
As agents take over building and validation, the MDLC reorganizes around evidence and approval. Humans set direction and accept evidence at gates. Policy, knowledge and skills become versioned infrastructure. Thesis v0 →
Research areas
Core themes
| # | Theme | Working question | Primary pages |
|---|---|---|---|
| T1 | Evidence as the primary artifact | What evaluation evidence is enough to approve a model or agent? | Evaluation |
| T2 | Human–AI responsibility | Which decisions stay human, and how are they recorded? | Responsibility |
| T3 | Supporting layers as infrastructure | How are policy, knowledge, skills and tools governed? | Layers |
| T4 | Convergence of ML and LLM lifecycles | Where do classical ML and agentic lifecycles diverge? | MDLC |
| T5 | AI coding environments | What is the right unit of reuse: skill, agent or workflow? | Workflows |
Architecture
Benchmarks
Key frameworks
Vendor landscape
| Vendor | Category | Build | Validate | Deploy | Monitor | Governance |
|---|---|---|---|---|---|---|
| Anthropic | Frontier lab | ? | ? | ? | ? | ? |
| OpenAI | Frontier lab | ? | ? | ? | ? | ? |
| Hyperscaler · lab | ? | ? | ? | ? | ? | |
| AWS | Hyperscaler | ? | ? | ? | ? | ? |
| Microsoft | Hyperscaler | ? | ? | ? | ? | ? |
| Open source | Ecosystem | ? | ? | ? | ? | ? |