Content
Working well with AI agents.
Why structure beats prompting, what good review and QA look like when an agent writes the code, and how Nanostack puts it into practice. Written from what we run, not what we imagine.
Your AI agent needs a workflow, not better prompts
Why AI-written code drifts out of scope, and how a phase-based workflow with saved artifacts fixes it without slowing the agent down.
Read →Workflow
What is a plan for, when the code takes five minutesWhy AI agents forget, and what to write downScope creep is the default mode of AI agentsQuality
QA that opens the app and tries itCode review for AI-written code, without reading every lineWhy our commits get blocked on purposeSafety
Prompt injection: when your AI agent reads hostile instructionsGuardrails for AI coding agents: what should never runAgentic security: the attack surface you installed last monthA security audit on every AI change, not once a quarterWorking with AI
How to read the checkpoints your AI agent leavesFrom vibe coding to something you can put in front of usersYou don't need to know how a code review worksFramework
When the default sprint is not your processRunning multiple AI coding agents on one repoEverything here describes the released product. When something only works on one host, the piece says so. Install nanostack →