Skip to content
View roy-tong's full-sized avatar

Achievements

Achievement: StarstruckAchievement: Quickdraw

Achievements

Achievement: StarstruckAchievement: Quickdraw

Block or report roy-tong

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
roy-tong/README.md

ROY TONG · Product Manager · Repeat Founder — turning frontier tech into shippable products

Roy Tong

Building AgentMeasure — open measurement & conformance infrastructure for AI agents.

Measure what agent telemetry actually counts: attempts, operations, retries, token accounting, and evidence boundaries.

AgentMeasure · Conformance vectors · Website · 中文主页

Recent external evidence

  • ✓ OpenLIT — token-accounting invariant merged upstream (reasoning ⊂ output; adding them reports tokens that never happened)
  • ✓ Urusilla — 2 external conformance vectors → 3 defects found and fixed, each with the finder's fixture as the regression guard
  • ✓ Agent semantics reviewed in public across Langfuse · DeepEval · OTel GenAI

Building now — AgentMeasure

AgentMeasure tests whether agent telemetry metrics actually mean what their labels claim.

  • Attempts vs operations — one logical use vs N executions
  • Retry accounting — a retried call is not two users
  • Token double counting — subsets must not be added into totals
  • Evidence vs inference — what a trace proves vs what it merely suggests
  • Eval repeatability — n runs are n measurements, not one verdict

Proven in the wild: an invariant merged into OpenLIT's main, two external conformance vectors from an independent contributor, three externally found defects fixed publicly. Long term: a common measurement layer for the Agent Capability Economy.

Try the 2-minute local demo · Send a trace, get a measurement check

Research

Product and industry research in four directions. Facts first; conclusions revised as evidence changes.

Direction Central question
Embodied Intelligence how robots move from demo to real tasks, product definition, and commercialization
Multimodal Interaction what the next interaction paradigm looks like when voice, vision, and space become I/O
On-device AI how models, compute, privacy, and form factors jointly shape the AI product experience
Agents how agents acquire context, choose capabilities, call software — and form a new software economy

Research library →

Writing

Judgments worth keeping, discussing, and falsifying (中文):

All essays →

Other tools

Tool What it does
User Demand Research (SURE) auditable demand evidence and research reports from goals, scope, and sample design (Skill + CLI + MCP)
iRead Research Monitor continuous source discovery with evidence-aware daily / weekly / monthly digests
Bilibili Video to Transcript public videos → timestamped, searchable research text
Roy's AI Product Research Library public research, essays, and an agent-readable index

Machine-readable tool catalog (agent routing): https://raw.githubusercontent.com/roy-tong/roy-tong/main/agent-tools.json

Router skill:

gh skill preview roy-tong/roy-tong find-research-tool
gh skill install roy-tong/roy-tong find-research-tool --agent codex --scope user

Product manager and repeat founder, ~10 years across AI, software, and intelligent hardware. About · GitHub · X / Twitter · Contact

Pinned Loading

  1. AgentMeasure AgentMeasure Public

    Open measurement infrastructure for agent-facing software. Separate operations from attempts, evidence from inference, and usage from value. Reach → Choice → Use → Utility → Value.

    Python 54

  2. iRead iRead Public

    iRead Research Monitor — Agent Skill and local-first CLI for source discovery, RSS or authorized WeChat collection, and evidence-aware recurring digests.

    Python 1

  3. roy-tong.github.io roy-tong.github.io Public

    Roy's AI Product Research Library — public knowledge base, llms.txt index, and read-only Agent Skill for agents, embodied AI, hardware, and commercialization.

    HTML 1

  4. user-demand-research user-demand-research Public

    User Demand Research (SURE) — Agent Skill for auditable voice-of-customer analysis, E0-E5 evidence grading, and opportunity validation.

    Python

  5. bilibili-transcript-pipeline bilibili-transcript-pipeline Public

    Bilibili Video to Transcript — Agent Skill and local CLI for Whisper transcription, timestamped Markdown, SRT subtitles, and structured JSON.

    Python