zxpmail (@tepeu) | Hashnode

@tepeuxian,chinaJoined July 2026

I build things that run in production. My writing focuses on AI agents, developer tooling, and the unsexy details of making unreliable systems behave predictably. All experiments and scripts are open-sourced. No claims without data.

@tepeuxian,chinaJoined July 2026

zxpmail

About

Available for

zxpmail's blogs

Recently published

The Red Line Principle: objective stop signals outperform LLM self-judgment in verifiable tasks

Six experiments on adversarial verification — and the 75% wall that didn't move

An alternative to LLM quality gates: deterministic routing + sampling

I designed a Harness to fix my agent's quality problem — then found 6 flaws in my own design

I tested 3 models as AI agent quality inspectors: the stronger the model, the more valid work it rejects

zxpmail

About

Available for

zxpmail's blogs

Recently published

The Red Line Principle: objective stop signals outperform LLM self-judgment in verifiable tasks

Six experiments on adversarial verification — and the 75% wall that didn't move

An alternative to LLM quality gates: deterministic routing + sampling

I designed a Harness to fix my agent's quality problem — then found 6 flaws in my own design

I tested 3 models as AI agent quality inspectors: the stronger the model, the more valid work it rejects

zxpmail

About

Available for

zxpmail's blogs

Recently published

The Red Line Principle: objective stop signals outperform LLM self-judgment in verifiable tasks

Six experiments on adversarial verification — and the 75% wall that didn't move

An alternative to LLM quality gates: deterministic routing + sampling

I designed a Harness to fix my agent's quality problem — then found 6 flaws in my own design

I tested 3 models as AI agent quality inspectors: the stronger the model, the more valid work it rejects

Search Hashnode

zxpmail

About

Available for

zxpmail's blogs

Recently published

The Red Line Principle: objective stop signals outperform LLM self-judgment in verifiable tasks

Six experiments on adversarial verification — and the 75% wall that didn't move

An alternative to LLM quality gates: deterministic routing + sampling

I designed a Harness to fix my agent's quality problem — then found 6 flaws in my own design

I tested 3 models as AI agent quality inspectors: the stronger the model, the more valid work it rejects