A practical AI agent development runbook covering task scope, isolated execution, six review gates, regression tests, bounded rework, and release evidence.
Badges
Badges come from GitHub once the author claims this feed.
Recent posts
every link opens the author’s post-
AI agent development: a runbook from brief to reviewed PR
-
Agentic SDLC: Approval, Recovery, and Evidence Contracts
An agentic SDLC needs version-bound approvals, safe recovery after uncertain writes, and independent evidence, without turning small changes into paperwork.
-
Designing a software factory for verifiable delivery
A software factory design with durable evidence, risk-based reviews and human approval, tested first on whether it stops safely rather than how fast it runs.
-
我想建的软件工厂:从需求到可验证交付
软件工厂不只是让智能体更快地写代码,而是把请求转化为可验证、可交接、可恢复的变更:本文梳理工单与执行的区别、持久证据、风险分级、独立审查和人工授权,讨论重复事件、执行中断与定向返工,并提出一个先测试失败时能否停止、再衡量真实交付成本的小规模试点方案,明确区分尚待检验的设计与已经验证的能力。