I write down judgments — and keep a public record of the ones new evidence changes.
This is my personal blog, about AI, embodied intelligence, products, and building companies: what technology can do, why people need it, and the choices involved in making it useful.
10 essays in English · site updated 2026-10-02
A few essays on the questions I keep coming back to.
New ways of finding, choosing, and using software call for new ways of measuring it.
How products change when intent, tasks, and outcomes become central.
Looking at robots through the abilities and reach they give people.
How an appointed manager earns trust and builds cooperation.
Each line is a current stance and the evidence beneath it. Stances move with evidence, and the revisions are logged in the open knowledge base.
AgentMeasure, open source, with public evidence cases
Seats, installs, and pageviews are breaking
Vision, presence, action, and physical reach
Geek, professional, B2B, consumer as evidence environments
Recent essays available in English. The Chinese archive has more.
v0.2.0 → v0.2.2 — an open experiment engine, the first preregistered A/B on a real agent, the first external fixture, the first public evidence case, and an honest 0%.
A baseline issue, not a news roundup — fixing the current state, the measurement method, and the data definitions that every future issue will compare against.
A field audit of six real claims, and what it means for the agent economy
Seats, installs, and pageviews are breaking. Measurement is the missing infrastructure of the AI economy.
Some questions become essays. Others become tools.
Agent measurement and conformance
Check whether attempts, logical operations, and retries are counted consistently in agent telemetry. Each check reports PASS, FAIL, or UNPROVABLE when evidence is missing.
A related token-accounting fix was merged into OpenLIT. Token checks are not yet supported by the Conformance Pack.
Ongoing reading and source discovery
Turn a research domain into a source collection and recurring daily, weekly, and monthly digests with sources and evidence boundaries.
Auditable user-demand research
Connect goals, scope, and sample design to demand evidence and research reports, preserving sources, evidence grades, and counterevidence. Available through Skill, CLI, and MCP interfaces.
For nearly a decade, I have built products across AI, software, and intelligent hardware, with several startup chapters along the way. This blog records my observations and how new evidence changes my mind.