Collect first agent research and market sources

2026-07-08 12:08:24 +08:00
parent 36c2499d3c
commit 1cbc98e432
34 changed files with 3417 additions and 58 deletions
@@ -0,0 +1,73 @@
+# Collection Run: 2026-07-08 First Sweep
+
+status: completed
+
+## Scope
+
+- date: 2026-07-08
+- jobs: rolling 6-month window, Beijing-first Agent / LLM roles
+- papers: provisional source sweep from arXiv and official research/blog sources
+- industry: large-company Agent work, technical reports, engineering blogs, benchmarks
+
+## Queries Used
+
+Jobs:
+
+```text
+site:joinbytedance.com 北京 Agent 大模型 LLM 招聘
+site:careers.tencent.com 北京 Agent 大模型 LLM 招聘
+site:talent.baidu.com 北京 Agent 大模型 LLM 招聘
+site:careers.alibaba.com 北京 Agent 大模型 LLM 招聘
+DeepSeek 招聘 Agent 北京 大模型 岗位
+site:jobs.bytedance.com 北京 智能体 大模型 Agent
+site:careers.tencent.com/jobdesc.html 北京 LLM Agent 桌面 智能伙伴
+```
+
+Papers:
+
+```text
+2026 LLM agent evaluation benchmark arxiv tool use memory
+2026 AI agent benchmark evaluation arxiv long horizon tasks
+2026 coding agent benchmark SWE-bench arxiv agent evaluation
+2026 multi-agent AI safety benchmark arxiv
+```
+
+Industry:
+
+```text
+site:openai.com/index agents 2026 technical report
+site:anthropic.com/news agent computer use 2026
+site:deepmind.google/blog agent 2026
+site:microsoft.com/en-us/research/blog agent 2026
+site:qwen.ai/blog agent Qwen 2026
+site:seed.bytedance.com/en agent 2026 ByteDance Seed
+site:github.com/deepseek-ai agent DeepSeek 2026
+```
+
+## First-pass Method Notes
+
+- Official pages are preferred for item files.
+- Search snippets are allowed only as weak evidence when official pages are JavaScript-rendered and not easily extractable.
+- For JD, source quality and location must be explicit; non-Beijing roles can be kept only when they reveal strong Agent demand signals.
+- For papers, items are marked `queued` or `skimmed`; source approval remains separate from item relevance.
+- For industry material, `source_quality` and `evidence_level` are deliberately separate.
+
+## Output
+
+- jobs: 6 starter items
+- papers: 6 starter items
+- industry: 9 starter items
+- scripts: item creation, index generation, URL checking
+
+## Summary
+
+- total items: 21
+- by collection: jobs 6, papers 6, industry 9
+- strongest recurring topics: agent evaluation, memory, coding agents, safety, enterprise deployment
+- first tool output: `data/index.json` and `data/summary.json`
+
+## Follow-up
+
+- Deepen official JD extraction for DeepSeek, ByteDance and Tencent.
+- Confirm approved paper sources before doing a larger paper sweep.
+- Add Anthropic and more Qwen/OpenAI/DeepSeek industry items in the next run.