# Roy.Tong — Writing and Open-source Projects

> Roy.Tong studies how emerging technology becomes real products, systems, and markets. His current campaign is the Agent Capability Economy: how software is discovered, chosen, invoked, metered, and paid for when software consumers shift from humans to agents. This index covers his public writing and open-source projects. Content is analysis unless a page links to primary evidence.

## Start Here

- [Home](https://roy-tong.github.io/): Concise introduction, recent writing, open-source projects, and contact.
- [Personal Knowledge Library](https://roy-tong.github.io/knowledge/): Six connected themes; 20 evolving knowledge notes, 36 article pages (including 10 English translations), and four open-source projects as of 2026-08-31. Original writing URLs remain stable.
- [Full-text knowledge index](https://roy-tong.github.io/knowledge/index.json): Public-only JSON with type, topic, language, provenance links, related notes, summaries and full text. Translations are not independent evidence.
- [Research data and limits](https://roy-tong.github.io/knowledge/data/): Reviewed aggregate counts for three private research snapshots. Records are not people or validated demands; raw platform content is not redistributed.
- [Knowledge source ledger](https://roy-tong.github.io/knowledge/sources/): Public primary sources with source roles and claim boundaries.
- [Reading and maintenance policy](https://roy-tong.github.io/knowledge/method/): Distinguish fact, author judgment, hypothesis, historical context, and revisions.
- [About Roy.Tong](https://roy-tong.github.io/about/): Public professional background and product experience.
- [Article archive](https://roy-tong.github.io/archive/): All published notes, ordered for human browsing.
- [Agent Capability Economy (Current Campaign)](https://roy-tong.github.io/capability-economy/): The current research campaign — how software is discovered, chosen, invoked, metered, and paid for when software consumers shift from humans to agents. Includes the thesis, the Research/Essays/Projects matrix, measurement terminology, and the Agent Capability Monthly structure.
- [Agent Capability Index](https://roy-tong.github.io/capability-index/): A public map of the agent software world — agent-invocable software by capability, with provider, interface, pricing, availability, and sources. Seed data, open for contribution.
- [Series A: Tech → Product Translation](https://roy-tong.github.io/series/tech-to-product/): Market roles, competitive analysis, demand evidence, and AI-native product form — reusable judgment methods for turning technology into product and business decisions.
- [Series B: Agents and AI Terminals](https://roy-tong.github.io/series/agent-and-terminals/): Agent runtimes, embodied intelligence, robots, AI wearables, and spatial computing — new product forms in information space and the physical world.
- [Series C: Product Management from Software to Hardware](https://roy-tong.github.io/series/software-to-hardware-pm/): Organization, stage gates, and system ownership across software and hardware.
- [RSS feed](https://roy-tong.github.io/feed.xml): Machine-readable updates.
- [Sitemap](https://roy-tong.github.io/sitemap.xml): Complete public URL inventory.

## Research Hubs

Long-form topic hubs under the Research program. Each hub tracks current change, market map, companies, products, technology paths, and commercialization evidence, with labeled facts, judgments, and open questions, plus per-hub update logs.

- [具身智能研究专题](https://roy-tong.github.io/research/embodied-intelligence/): Embodied-intelligence hub — mass-production year one, market layers, company and product tables, VLA/world-model technology paths, commercialization evidence, and Roy's research materials and essays.
- [Embodied Intelligence Hub (EN)](https://roy-tong.github.io/en/research/embodied-intelligence/): English edition of the embodied-intelligence hub.
- [Research section](https://roy-tong.github.io/research/): Existing research programs and the six-theme knowledge library. Capability Economy and the original embodied-intelligence hub retain their canonical URLs.

## Connected Knowledge Themes

- [Agent and software systems](https://roy-tong.github.io/knowledge/topics/agent-systems/): Runtime context, callable capabilities, measurement, and research workflows.
- [Embodied intelligence and robotics](https://roy-tong.github.io/knowledge/topics/robotics/): Industry structure, household tasks, manipulation data, world models, and delivery evidence.
- [AI devices and interaction](https://roy-tong.github.io/knowledge/topics/ai-devices/): Local AI, wearables, spatial objects, permissions, and actual usage conditions.
- [Creative tools and digital manufacturing](https://roy-tong.github.io/knowledge/topics/creative-tools/): Editable video projects, persistent interactive states, and physical production.
- [Product research and business judgment](https://roy-tong.github.io/knowledge/topics/product-research/): Demand evidence, alternatives, market entry, denominators, and revision.
- [Organization and industrial execution](https://roy-tong.github.io/knowledge/topics/organization/): Irreversible hardware investment, responsibility interfaces, and capability transfer.

## Agent Tools and Components

- [Agent Research Toolkit catalog](https://raw.githubusercontent.com/roy-tong/roy-tong/main/agent-tools.json): Machine-readable task routing, interfaces, installation commands, first-success checks, constraints, and privacy boundaries for all public tools below.
- [iRead Research Monitor](https://github.com/roy-tong/iRead): Local-first continuous research monitor with source discovery and evidence-aware daily, weekly, and monthly reports.
- [User Demand Research (SURE)](https://github.com/roy-tong/user-demand-research): An installable Agent Skill, a deterministic CLI, and a stdlib-only MCP server for stage-gated demand research — E0–E5 evidence, auditable demand judgments, a plan/signals/report pipeline, and a machine-readable registry that reviews GitHub OSS connectors for Reddit, X, YouTube, Amazon, JD, Taobao/Tmall, and Kickstarter. Agent index: https://raw.githubusercontent.com/roy-tong/user-demand-research/main/llms.txt
- [Bilibili Video to Transcript](https://github.com/roy-tong/bilibili-transcript-pipeline): Local video-to-Markdown, SRT, and JSON pipeline with an offline first-success demo, Agent Skill, and CLI.
- [AgentMeasure](https://github.com/roy-tong/AgentMeasure): Open measurement infrastructure for the Agent Capability Economy — an open standard (canonical observations, evidence-graded identity, metrics registry) plus an MCP SDK for measuring how agents discover, choose, use, and derive value from software capabilities. Website with trial CTA and blog: https://roy-tong.github.io/AgentMeasure/. Current state (v0.2.2): conformance vector suite with the first community-contributed fixture (Urusilla-001), the first public evidence case (langfuse demo traces — 12 attempts, 0% grouping evidence, 0% safe operation coverage, recomputable), the open-source Lab experiment engine (preregistered A/Bs, honest statistics, 78 tests) live-validated on a real agent harness, and a two-state consumption ontology (availability / influence, DR-005).
- [Roy's AI Product Research Library](https://github.com/roy-tong/roy-tong.github.io/tree/main/skills/research-knowledge-base): Read-only Agent entry point for this public site.

## Current Release Sequence

- [AgentMeasure 一周进展：规范开始长出证据](https://roy-tong.github.io/notes/agentmeasure-from-spec-to-evidence/): Progress report for Aug 22–25 — Lab v0.4 open experiment engine, the first preregistered A/B on a real agent (honest null, real token metering), v0.2.2 conformance hardening with the first community fixture (Urusilla-001), the first public evidence case (12 attempts, 0% safe operation coverage, recomputable), and the spec's move from a consumption ladder to two states plus ranked evidence.
- [AgentMeasure Weekly: The Spec Is Growing Evidence (EN)](https://roy-tong.github.io/en/notes/agentmeasure-from-spec-to-evidence/): English edition of the progress report.
- [AI 应用不会被 Agent 塞满，它们会先被拆掉](https://roy-tong.github.io/notes/ai-apps-unbundled-before-agent-filled/): DeepSeek Harness and Codex Harness point to a route more radical than "adding agents to software" — model, skills, tools, data, runtime, and UI composed temporarily at runtime and dissolved after the task. The basic unit of software shifts from App to Capability; SaaS gets unbundled; the business model moves from Seat Economy to Machine Economy.
- [The Upsider 事件：Agent 第一次作为经济主体收付了钱](https://roy-tong.github.io/notes/the-upsider-event/): An AI agent on X evaluated interactions and automatically sent tokens on-chain — settlement is verifiable (E4) while the semantics of reward and value attribution are undefined. Payment is verifiable; value attribution isn't. Agents are becoming economic actors.
- [Agent Capability Monthly · Issue 01: 生态基线](https://roy-tong.github.io/notes/agent-capability-monthly-01/): The baseline issue of the fixed monthly report on the Agent Capability Economy — 106 agent-invocable capabilities across 11 categories, pricing-war implications, MCP Registry consolidation, agent commerce protocols (ACP/AP2/x402/Bazaar), and why payment companies building metering validates "measurement precedes payment."
- [Agent Capability Monthly · Issue 01 (EN)](https://roy-tong.github.io/en/notes/agent-capability-monthly-01/): English edition of the baseline issue.
- [空间计算需要自己的“触控时刻”](https://roy-tong.github.io/notes/spatial-computing-touch-moment/): A public interaction grammar for spatial computing: look, point, pinch, and speak; Point and Ask / Point and Act; fast direct manipulation and slower Agent execution.
- [每个 Agent 用量数字，都是自报的](https://roy-tong.github.io/notes/every-agent-usage-number-is-self-reported-zh/): A field audit of six real agent-usage claims (x402, llms.txt audit, ClaudeBot impersonation, registry badges), graded on an evidence ladder — every number is self-reported; independence beats volume; nobody publishes the unit definition. Companion to AgentMeasure Benchmark Run #001.
- [Every Agent Usage Number Is Self-Reported (EN)](https://roy-tong.github.io/en/notes/every-agent-usage-number-is-self-reported/): English edition of the claim audit essay.
- [RAG 之后，Agent 需要 Context Recommendation](https://roy-tong.github.io/notes/agent-context-recommendation-after-rag/): A proposal for an Agent Runtime layer that assembles the information, capabilities, state, and authority each step should receive, informed by DeepSeek Harness.
- [家庭机器人为什么技术上更难，经济上可能更丰富](https://roy-tong.github.io/notes/home-robots-harder-richer/): Why household long tails and multi-person rules raise technical difficulty while hardware, subscriptions, services, care, and coordination create additional value and responsibility layers.
- [机器人不是替代人，而是给人增加新的身体](https://roy-tong.github.io/notes/robots-give-people-new-bodies/): A capability-gain framework for robots that extend human vision, presence, action, and physical attributes through shared control.
- [空降管理者如何从职位合法性走向组织合法性](https://roy-tong.github.io/notes/appointed-manager-organizational-legitimacy/): How an appointed leader moves from formal authority to organizational legitimacy through stabilize-and-diagnose, shared evidence, decision trust, and fair change.
- [Geek、Professional、B、C：不同市场负责消灭不同不确定性](https://roy-tong.github.io/notes/market-roles-remove-uncertainty/): A market-entry framework that treats Geek, Professional, domain B2B, and Consumer users as evidence environments rather than a fixed maturity ladder.

## AI Products and Agents

- [当软件的消费者变成 Agent](https://roy-tong.github.io/notes/when-the-software-consumer-becomes-an-agent/): The flagship AgentMeasure essay — why the old measurement chain (install, seat, pageview) breaks for agent consumers, why measurement precedes payment, and what an open, verifiable measurement layer for the Agent Capability Economy looks like.
- [竞品分析为什么不该从参数表开始](https://roy-tong.github.io/notes/competitive-analysis-software-hardware/): Why software competition must be compared through workflow, data, migration, and continuous delivery, while hardware also requires physical-system, manufacturing, reliability, channel, and service evidence.
- [AI Native 之后，产品的基本单位变了](https://roy-tong.github.io/notes/ai-native-basic-unit/): Why semantic objects, tasks, relations, and operation history may become the primary state of AI-native products while pages, files, and editors become views and control surfaces.
- [Agent 向左，具身向右：AI 在信息空间与物理世界的分岔](https://roy-tong.github.io/notes/agent-left-embodied-right/): Why Agent products and embodied systems share model capabilities but diverge in reversibility, evaluation, safety, deployment, and economics.
- [WAIC之后：AI产业开始为“把事做成”买单](https://roy-tong.github.io/notes/waic-from-models-to-systems/): From model capability to systems, tools, workflow, delivery, and organizational adoption.
- [如何从 20 万条反馈中形成可验证的需求判断](https://roy-tong.github.io/notes/scene-user-demand-evidence-research/): A practical human guide to SURE, including decision design, GitHub OSS connector review, platform source routes, evidence records, audit gates, and product validation.

## Embodied AI and Intelligent Hardware

- [AI Wearable 的竞争，不是眼镜对耳机](https://roy-tong.github.io/notes/ai-wearable-modalities-body-comfort/): A four-layer framework for comparing AI wearables through input-output modality combinations, multimodal controls, context-dense body positions, and long-term wearing cost rather than glasses-versus-earbuds categories.
- [具身智能入门：产业、公司、产品、技术与职业地图](https://roy-tong.github.io/notes/embodied-intelligence-beginners-guide/): A structured beginner map of the embodied-intelligence field.
- [从 AI 应用到真实世界：我的一次转向](https://roy-tong.github.io/notes/from-ai-software-to-physical-world/): Why AI product work shifts from software-only interaction toward physical-world tasks.
- [智能硬件创新，为什么越来越依赖组织能力](https://roy-tong.github.io/notes/hardware-innovation-organization-management/): Product and organizational constraints behind shipping intelligent hardware.

## Product Practice and Writing

- [重新开始写作](https://roy-tong.github.io/notes/start-writing/): Why this public research archive exists and how judgments will be revised over time.

## English Writing

- [When the Software Consumer Becomes an Agent](https://roy-tong.github.io/en/notes/when-the-software-consumer-becomes-an-agent/): Flagship AgentMeasure essay (English) — measurement before payment, the five-link chain, evidence discipline, and why the measurement layer must be an open standard.
- [Geek, Professional, B2B, Consumer: Markets as Evidence Environments](https://roy-tong.github.io/en/notes/market-roles-remove-uncertainty/): English edition of the market-entry framework — Geek, Professional, domain B2B, and Consumer users as evidence environments rather than a maturity ladder.
- [After AI-Native: The Basic Unit of a Product Has Changed](https://roy-tong.github.io/en/notes/ai-native-basic-unit/): English edition — why semantic objects, tasks, relations, and operation history become the primary state of AI-native products.
- [After RAG, Agents Need Context Recommendation](https://roy-tong.github.io/en/notes/agent-context-recommendation-after-rag/): English edition — an Agent Runtime layer that assembles the information, capabilities, state, and authority each step should receive.
- [Robots Don't Replace People — They Give People New Bodies](https://roy-tong.github.io/en/notes/robots-give-people-new-bodies/): English edition — a capability-gain framework for robots that extend human vision, presence, action, and physical attributes.
- [From Formal Authority to Organizational Legitimacy](https://roy-tong.github.io/en/notes/appointed-manager-organizational-legitimacy/): English edition — how an appointed leader moves from formal authority to organizational legitimacy.

## Usage Notes

- Cite the original article URL, not this index, when using an argument.
- Check publication context and newer evidence before treating an article as current fact.
- The site does not expose private notes, user data, credentials, or unpublished Feishu content.
- [AI 视频软件的护城河，藏在第二次修改里](https://roy-tong.github.io/notes/ai-video-second-edit/): Why preserving approved state through successive edits may matter more than first-generation demos in professional creative workflows; author thesis, not a vendor benchmark.
- [本地 AI 盒子，得先证明自己值得多占一个插座](https://roy-tong.github.io/notes/local-ai-box-task-economics/): Evaluate a dedicated device against existing computers and allowed alternatives using full task cost, privacy boundaries, uptime, and maintenance.
- [机器人进家庭，要先过“谁来收拾残局”这一关](https://roy-tong.github.io/notes/home-robots-recovery-burden/): Evaluate household preparation, recovery, intervention and maintenance; buyers, users and maintainers may be different people.