Add industry collection and six-month job tracking
This commit is contained in:
wuyang
@@ -0,0 +1,35 @@
|
||||
# Industry
|
||||
|
||||
这里持续收集大公司和重要 AI 团队的 Agent 相关工作：产品发布、技术报告、研究博客、工程博客、代码仓库、benchmark 和实践复盘。
|
||||
|
||||
它和 `papers/` 的区别：`papers/` 关注研究证据，`industry/` 关注公司真实在做什么、怎么落地、暴露了哪些工程和产品趋势。
|
||||
|
||||
## Structure
|
||||
|
||||
- [items](items/README.md): 每条公司工作一个 Markdown 文件
|
||||
- [schema](schema.md): 行业资料字段和写法
|
||||
- [source-registry](source-registry.md): 官方来源和可信来源清单
|
||||
- [company-watchlist](company-watchlist.md): 重点公司和团队
|
||||
- [industry-insights](industry-insights.md): 周期性趋势总结
|
||||
|
||||
## Collection Questions
|
||||
|
||||
- 大公司正在把 Agent 用在哪些产品和场景？
|
||||
- 它们公开了哪些系统设计、工具调用、记忆、评估、安全和部署经验？
|
||||
- 哪些工作只是产品宣传，哪些有技术细节或可复现材料？
|
||||
- 哪些行业实践能反哺我们的项目、实验和知识文档？
|
||||
- 公司实践和 JD 要求、论文趋势之间是否互相印证？
|
||||
|
||||
## Workflow
|
||||
|
||||
1. 收集官方博客、技术报告、产品文档、代码仓库或演讲资料。
|
||||
2. 用 [industry-note](../templates/industry-note.md) 建立结构化笔记。
|
||||
3. 标注 company、source_type、topics、implementation_signals 和 evidence_level。
|
||||
4. 更新 [industry-insights](industry-insights.md)。
|
||||
5. 把稳定结论回流到 `docs/`、`papers/`、`jobs/`、`experiments/` 或 `projects/`。
|
||||
|
||||
## Initial Scope
|
||||
|
||||
- frontier labs: OpenAI、Anthropic、Google DeepMind、Microsoft、Meta、Amazon、NVIDIA
|
||||
- China AI teams: DeepSeek、腾讯、字节 Seed、阿里 Qwen、百度、智谱、月之暗面
|
||||
- product areas: coding agent、computer use、deep research、office agent、enterprise agent、agent safety、multi-agent、agent evaluation
|
||||
@@ -0,0 +1,31 @@
|
||||
# Company Watchlist
|
||||
|
||||
## Frontier Labs
|
||||
|
||||
| Company | Focus | Priority |
|
||||
| --- | --- | --- |
|
||||
| OpenAI | Codex、computer use、deep research、agent productization、evaluation | P0 |
|
||||
| Anthropic | Claude Code、computer use、tool use、安全、enterprise agent | P0 |
|
||||
| Google DeepMind | Gemini agent、multi-agent safety、agent security、research reports | P0 |
|
||||
| Microsoft | Copilot、Agent AI、software engineering agents、enterprise agent ecosystem | P0 |
|
||||
| Meta | Llama、open models、agent frameworks、consumer AI | P1 |
|
||||
| Amazon | Alexa、AWS agent services、industrial AI applications | P1 |
|
||||
| NVIDIA | Agent systems、inference、robotics、accelerated deployment | P1 |
|
||||
|
||||
## China AI Teams
|
||||
|
||||
| Company / Team | Focus | Priority |
|
||||
| --- | --- | --- |
|
||||
| DeepSeek | reasoning、coding、open models、agent capabilities、infra | P0 |
|
||||
| 字节 Seed | Doubao、Seed models、agent era、world models、AI infra | P0 |
|
||||
| 阿里 Qwen | Qwen Agent、coding、multimodal agents、open models | P0 |
|
||||
| 腾讯 | 混元、元宝、企业应用、游戏和内容场景、AI Lab | P0 |
|
||||
| 百度 | 文心、搜索、智能云、Agent 应用 | P1 |
|
||||
| 智谱 | GLM、Agent 平台、企业应用 | P1 |
|
||||
| 月之暗面 | Kimi、长上下文、AI 助手 | P1 |
|
||||
|
||||
## Review Rules
|
||||
|
||||
- P0 每周扫一轮重大更新。
|
||||
- P1 每月扫一轮，除非出现明显热点。
|
||||
- 新公司先进入候选，连续出现高价值 Agent 信息后再提升优先级。
|
||||
@@ -0,0 +1,53 @@
|
||||
# Industry Insights
|
||||
|
||||
status: seed
|
||||
|
||||
这个文件沉淀大公司 Agent 工作和技术报告的趋势判断。单条资料放在 `industry/items/`，这里记录跨公司的归纳。
|
||||
|
||||
## Current Snapshot
|
||||
|
||||
- date:
|
||||
- scope:
|
||||
- sources reviewed:
|
||||
|
||||
## Strong Industry Signals
|
||||
|
||||
-
|
||||
|
||||
## Weak or Marketing-heavy Signals
|
||||
|
||||
-
|
||||
|
||||
## Company Patterns
|
||||
|
||||
| Company | Agent Direction | Evidence | Notes |
|
||||
| --- | --- | --- | --- |
|
||||
| OpenAI | _TBD_ | _TBD_ | _TBD_ |
|
||||
| Anthropic | _TBD_ | _TBD_ | _TBD_ |
|
||||
| Google DeepMind | _TBD_ | _TBD_ | _TBD_ |
|
||||
| Microsoft | _TBD_ | _TBD_ | _TBD_ |
|
||||
| DeepSeek | _TBD_ | _TBD_ | _TBD_ |
|
||||
| 腾讯 | _TBD_ | _TBD_ | _TBD_ |
|
||||
| 字节 Seed | _TBD_ | _TBD_ | _TBD_ |
|
||||
| 阿里 Qwen | _TBD_ | _TBD_ | _TBD_ |
|
||||
|
||||
## Implementation Patterns
|
||||
|
||||
- tool use:
|
||||
- memory:
|
||||
- evaluation:
|
||||
- safety:
|
||||
- deployment:
|
||||
- productization:
|
||||
|
||||
## Follow-up Queue
|
||||
|
||||
-
|
||||
|
||||
## Links Back to Knowledge Base
|
||||
|
||||
- docs to update:
|
||||
- papers to read:
|
||||
- jobs to compare:
|
||||
- experiments to create:
|
||||
- projects affected:
|
||||
@@ -0,0 +1,17 @@
|
||||
# Industry Items
|
||||
|
||||
每条公司工作一个文件，文件名建议：
|
||||
|
||||
```text
|
||||
YYYY-MM-DD-company-short-title.md
|
||||
```
|
||||
|
||||
例子：
|
||||
|
||||
```text
|
||||
2026-07-08-openai-agents-transforming-work.md
|
||||
2026-07-08-google-deepmind-agent-security.md
|
||||
2026-07-08-qwen-agentworld.md
|
||||
```
|
||||
|
||||
如果同一公司连续发布多个相关材料，优先拆成多条 item，再在 `industry-insights.md` 中做汇总。不要把多个来源混成一条笔记，否则后续很难追踪证据。
|
||||
@@ -0,0 +1,105 @@
|
||||
# Industry Schema
|
||||
|
||||
行业资料笔记使用 YAML frontmatter 加正文。
|
||||
|
||||
## Frontmatter
|
||||
|
||||
```yaml
|
||||
---
|
||||
type: industry
|
||||
company:
|
||||
team:
|
||||
title:
|
||||
url:
|
||||
source_name:
|
||||
source_type: technical-report / research-blog / engineering-blog / product-blog / product-doc / code-release / benchmark / talk / other
|
||||
source_quality: official / official-adjacent / secondary / unknown
|
||||
published_at:
|
||||
collected_at: YYYY-MM-DD
|
||||
status: queued / summarized / analyzed / deprecated
|
||||
topics:
|
||||
- agent
|
||||
implementation_signals:
|
||||
- tool-use
|
||||
product_area:
|
||||
- coding-agent
|
||||
models:
|
||||
-
|
||||
tools:
|
||||
-
|
||||
benchmarks:
|
||||
-
|
||||
related_papers:
|
||||
-
|
||||
related_jobs:
|
||||
-
|
||||
related_experiments:
|
||||
-
|
||||
related_projects:
|
||||
-
|
||||
evidence_level: high / medium / low
|
||||
relevance: high / medium / low
|
||||
---
|
||||
```
|
||||
|
||||
## Body
|
||||
|
||||
```text
|
||||
# <company> - <title>
|
||||
|
||||
## One-line Takeaway
|
||||
|
||||
一句话说明这条资料暴露了什么行业趋势或工程经验。
|
||||
|
||||
## Source Snapshot
|
||||
|
||||
来源、发布日期、采集日期、来源质量。
|
||||
|
||||
## What They Built or Claimed
|
||||
|
||||
公司发布了什么产品、系统、能力、报告或代码？
|
||||
|
||||
## Technical Signals
|
||||
|
||||
- architecture:
|
||||
- tool use:
|
||||
- memory:
|
||||
- evaluation:
|
||||
- safety:
|
||||
- deployment:
|
||||
- data:
|
||||
|
||||
## Product and Business Signals
|
||||
|
||||
- users:
|
||||
- scenario:
|
||||
- pricing / availability:
|
||||
- go-to-market:
|
||||
|
||||
## Evidence Quality
|
||||
|
||||
判断它是强证据还是弱信号。是否有论文、技术报告、benchmark、代码或只是营销描述？
|
||||
|
||||
## Links
|
||||
|
||||
- papers:
|
||||
- jobs:
|
||||
- experiments:
|
||||
- projects:
|
||||
- docs:
|
||||
|
||||
## Gaps for Us
|
||||
|
||||
-
|
||||
|
||||
## Notes
|
||||
|
||||
-
|
||||
```
|
||||
|
||||
## Field Rules
|
||||
|
||||
- `url` 和 `collected_at` 必填。
|
||||
- 优先官方来源；二手来源只能作为线索。
|
||||
- `evidence_level` 和 `source_quality` 必须分开：官方产品博客也可能证据很弱。
|
||||
- 技术细节不足时标记为 `low`，不要强行推断实现。
|
||||
@@ -0,0 +1,67 @@
|
||||
# Industry Source Registry
|
||||
|
||||
这个文件记录大公司 Agent 工作、技术报告和工程博客的来源。
|
||||
|
||||
## Source Priority
|
||||
|
||||
| Priority | Source Type | Use |
|
||||
| --- | --- | --- |
|
||||
| 1 | 官方研究页、官方博客、官方技术报告、官方文档 | 直接整理 |
|
||||
| 2 | 官方 GitHub、Hugging Face、模型卡、benchmark 页面 | 直接整理，但注意版本 |
|
||||
| 3 | 官方会议演讲、开发者大会、产品发布页 | 可整理，标注产品宣传成分 |
|
||||
| 4 | 可信媒体访谈、第三方深度分析 | 作为线索或补充，不单独形成强结论 |
|
||||
|
||||
## Candidate Official Sources
|
||||
|
||||
这些来源可作为第一批候选。后续每次收集前可以按主题取子集。
|
||||
|
||||
| Company / Team | Source | URL | Notes |
|
||||
| --- | --- | --- | --- |
|
||||
| OpenAI | Research | https://openai.com/research/ | 研究论文、系统报告、agent 相关研究 |
|
||||
| OpenAI | News / Index | https://openai.com/index/ | 产品、经济研究、发布博客 |
|
||||
| Anthropic | Research | https://www.anthropic.com/research | 研究报告和安全材料 |
|
||||
| Anthropic | News | https://www.anthropic.com/news | 产品发布、computer use、Claude 相关工作 |
|
||||
| Google DeepMind | Blog | https://deepmind.google/blog/ | agent、安全、多 Agent、Gemini 相关博客 |
|
||||
| Google DeepMind | Publications | https://deepmind.google/research/publications/ | 论文和技术报告索引 |
|
||||
| Google Research | Blog | https://research.google/blog/ | 研究博客和系统实践 |
|
||||
| Microsoft Research | Blog | https://www.microsoft.com/en-us/research/blog/ | Agent、安全、软件工程和研究博客 |
|
||||
| Microsoft Research | Agent AI project | https://www.microsoft.com/en-us/research/project/agent-ai/ | Agent 项目页 |
|
||||
| Meta AI | Blog | https://ai.meta.com/blog/ | Llama、agents、产品和研究发布 |
|
||||
| Amazon Science | Home | https://www.amazon.science/ | 工业研究和应用论文 |
|
||||
| NVIDIA Research | Home | https://research.nvidia.com/ | Agent、模型、系统研究 |
|
||||
| NVIDIA Technical Blog | Blog | https://developer.nvidia.com/blog/ | 工程博客、部署和加速实践 |
|
||||
| DeepSeek | Home / News | https://www.deepseek.com/ | 官方入口和岗位入口 |
|
||||
| DeepSeek | GitHub | https://github.com/deepseek-ai | 模型、报告和代码发布 |
|
||||
| ByteDance Seed | Home | https://seed.bytedance.com/en/ | Seed 团队入口 |
|
||||
| ByteDance Seed | Research | https://seed.bytedance.com/en/research | 研究和技术发布 |
|
||||
| ByteDance Seed | GitHub | https://github.com/ByteDance-Seed | 代码和项目发布 |
|
||||
| Qwen | Blog | https://qwen.ai/blog | Qwen 模型、Agent 和技术博客 |
|
||||
| Qwen | Hugging Face | https://huggingface.co/Qwen | 模型卡、发布和代码入口 |
|
||||
| Alibaba Cloud | Blog | https://www.alibabacloud.com/blog | 阿里云工程博客和产品实践 |
|
||||
| Tencent AI Lab | Home | https://ai.tencent.com/ailab/ | 腾讯 AI Lab 研究入口 |
|
||||
| Baidu Research | Home | https://research.baidu.com/ | 百度研究入口 |
|
||||
|
||||
## Collection Queries
|
||||
|
||||
```text
|
||||
site:openai.com/index agent technical report
|
||||
site:anthropic.com/news computer use agent
|
||||
site:deepmind.google/blog agent safety multi-agent
|
||||
site:microsoft.com/en-us/research/blog agent evaluation
|
||||
site:qwen.ai/blog agent coding
|
||||
site:seed.bytedance.com/en agent LLM
|
||||
site:github.com/deepseek-ai agent
|
||||
site:ai.tencent.com/ailab agent large language model
|
||||
```
|
||||
|
||||
## Collection Cadence
|
||||
|
||||
- weekly: 新发布、重大产品或技术报告
|
||||
- monthly: 公司实践趋势和技术信号
|
||||
- quarterly: 行业实践和论文/JD/项目之间的交叉分析
|
||||
|
||||
## Quality Notes
|
||||
|
||||
- 产品博客通常能说明方向，但不一定能说明实现。
|
||||
- 技术报告、代码和 benchmark 的证据强度更高。
|
||||
- 公司公开信息天然带宣传倾向，需要和论文、代码、实际产品体验或 JD 互相校验。
|
||||
Reference in New Issue
Block a user