← All entries

Diary of an AI Practitioner

Building

Building

Instructions, tools and triggers, and mapping the work before you automate it.

7 entries

Standing Orders: How Agent Instructions Differ From a Prompt

A prompt is a one-time ask. Instructions are standing orders that live with the agent and apply every run. If you retype your rules every session, you have a habit, not an agent.

Giving an Agent Hands: What Tools Actually Change

An agent with no tools can only talk. Tools are access to the calendar, the files, the inbox, whatever the job needs. It is the line between advice and action, and it is where the real risk lives.

Where to Start With AI Agents: Pick One Real Job

Not a framework, not a whole system. One job, clear instructions, the one tool it needs, and a trigger so it runs without you. Then watch it before you trust it with more.

Triggers: What Starts an Agent When You Don't

A trigger is what starts the work without you typing anything. It is the actual line between using a convenient tool and having something work for you, and it is the piece most people skip.

Do It Twice, Then Write It Down as a System

Once is a task. Twice is a pattern. The third time you are about to do it, stop and write the steps down instead. You need three things first, and none of them is software.

Ask Your AI to Show Its Work, Not Just Do It

AI work is invisible, which is why it feels out of control. Ask the agent to show you the work, not just do it, and the whole thing becomes something you can actually manage.

How I Use Voice Notes and Brain Dumps to Think

My process is audio first, not outline first, and forcing myself into the standard advice would have broken the way I actually work.

— Building —