How we work with AI has been climbing for years. Prompt engineering, context engineering, steering engineering, loop engineering: these four stages form one spine. This map hangs all twenty-odd related articles I have written onto it. Enter at any station and just read down.
The further down you go, the less you manage how AI does it and the more you focus on "what you want." Click any card to start there.
Fine-tuning how you phrase a single instruction. This is the starting point and the first thing most people learn. A single-line prompt has a low ceiling, so the focus moves up quickly.
Managing the whole conversation: what data, background, and surrounding context you give AI. How you organize a knowledge base is the core of this stage.
When there is so much data that AI cannot find the point, how to split a knowledge base into three.
Turn Tags into Cards to Bring a Knowledge Base Alive: The Tag Wiki MethodUse controlled tags to string notes into a wiki so AI can read your knowledge.
Write a Diary, Put AI to Work: The 3x4 MethodKeep feeding AI context by writing diaries, and it can pick up your work.
AI Agents Are Taking Over Your Workflows: Are Your Processes and Knowledge Ready?Organize yourself into a format both people and AI can read, so agents can hand off between each other.
Give AI your intent and bottom line, and guide a collaborator smarter than you toward what you want. This stage has the most articles; it is the current home turf.
Why intent first, plus three entry moves and three copy-paste prompts.
Series 02 The Boss’s Steering Mindset: Say to AI What You Would Never Say to an EmployeeTreat AI as an employee, be the owner toward it, with the ten steering questions and inward questioning.
Docs as System Design: How Non-Engineers Build Agent FrameworksUse docs as system design so AI runs along the architecture you set.
Build Your AI Employees + ConsultantsLead AI the way you lead employees: verify first, then delegate.
I Caught My AI Slacking, Then Wrote It Into the RulesWhen AI messes up: distill your requirements into rules, and it will follow them next time.
Ponytail: The "Decision Ladder" for People Who Don’t Write CodeA judgment framework for steering AI: when to let go and when to rein in.
Where Do You Stand in the AI World? A 14-Tier Ranking by Industry ImpactHow smart AI is now and where you stand decide how you steer it.
Stop giving instructions one by one and design a loop that runs on its own. Turn your whole workflow into a loop that accumulates automatically, and can tolerate errors.
Uses one real workflow as an example to explain, in plain words, what loop engineering is.
When I Started Thinking in Loops: Three Workflows Redesigned as LoopsA publishing loop, a retrospective loop, a distillation loop: how to design three loops.
I Used an Agent to Build My Automated WorkflowHand repetitive work to an AI agent and grow it into an automation.
Building Your Own Automatic Work Log with Codex AloneLet AI log your work automatically every day, so you have context to look back on.
How to Auto-Archive Data Scattered Across LINE Groups Every DayA loop that runs itself daily, pulling group messages and files into your knowledge base.
My Agent-Powered Meeting Workflow: Seven Steps to Turn Meetings into Knowledge AssetsFrom recording to finished draft, a repeatable meeting loop.
AI Errors Aren’t the Threat, No Backup Is: Lessons from Breaking 170 FilesA self-running loop must have backups, so you can come back when it goes wrong.
Rather Than a Flawless AI System, Design a Loop That RecoversThe point of the design is not avoiding errors, but recovering from them on its own.