How We Taught Our Codex Workflow Anthropic's Principles
How we reshaped our working system after studying Anthropic courses: planning, verification, hooks, skills and safe automation.
Read the article →PRACTICAL AI ENGINEERING
AI agents for real work
Field notes from a working lab: how we build, test, monitor and improve AI agents before trusting them with real processes.
Latest articles
Protecting AI Agent Tools with the Doberman MCP Proxy
A hands-on benchmark for policy enforcement, explicit outcomes and repeatable security checks.
Read the article →How We Built an Autonomous Publication Pipeline
What failed when articles, catalogs, feeds and publication checks drifted apart, and how we fixed the workflow.
Read the article →How an AI Agent Stops Being a Dumb Telegram Autoresponder
Routing, memory, approvals and tests for an agent that works with real messages.
Read the guide →What you will find here
- Repeatable workflows for email, Telegram, documents and research.
- Failure reports with the fix, verification steps and remaining limits.
- Clear explanations of agent memory, approvals, costs and security.