Catch the Behavior AI Quietly Changed, and Turn It Into a Workflow

Published: 2026-07-04Updated: 2026-07-04

What this piece is about

While checking my itinerary on a business trip, I noticed my first move had shifted from "opening Google Maps" to "asking AI," and I hadn't noticed at all. Starting from that case, this piece is about something I think everyone should do: find these behaviors AI has quietly changed, capture them as a concrete, well-defined workflow, and eventually even let AI run them for you. A copy-paste prompt is included at the end.

Who it's for

· People who use AI every day yet can't say exactly which work habits AI has changed
· People who often ask AI the same kind of question and have to describe it from scratch each time
· People who want AI to move from "answering questions" to "taking over the process"

What you'll get

· A way to spot the signal: how to notice something has quietly been handed to AI
· A four-step method following Loop Engineering: find the behavior, observe the variables, fix the process, make it a prompt
· A copy-paste prompt you can use directly

One image: the signal, the fork, and organizing it into a repeatable workflow

A business-trip case

Yesterday, on a business trip, I noticed one of my habits had been swapped out by AI.

When checking the itinerary, I didn't open Google Maps and drag out the route segment by segment. I handed the situation straight to AI: how the normal schedule goes, what to do if it ends an hour early, what to do if it ends an hour late, and asked it to find the routes and train times in advance. Google Maps was left with only one job: confirming the departures on site.

That's when I realized my first move for planning a trip had already shifted from opening the map to asking AI.

This is a signal

What makes this worth noticing is the skipping: I skipped a step I used to do by hand, and skipped it without realizing. Plenty of people use AI; the signal is that I quietly stopped doing that step myself.

Everyone has these moments. It could be looking things up, planning a schedule, writing an email, or putting together a report. One day you look back and realize you've already defaulted to asking AI first for that task.

What this meansThis is actually a signal worth pausing on: it means a piece of judgment you used to keep in your head has started being outsourced to AI. That piece of judgment is your own tacit knowledge; you just never wrote it down.

Here's a fork

Let it pass

You think "AI is so convenient," and next time the same thing comes up, you ask again on the spot. The convenience is one-off, and nothing is left behind.

Catch it

You organize the task into a concrete, well-defined workflow. An organized workflow accumulates, can be reused, and can be handed off, and eventually AI can even run it automatically.

The difference between the two paths is the difference between "AI answers you once" and "AI starts accumulating for you."

Following Loop Engineering, organize the behavior into a workflow AI can run repeatedly

Following the Loop Engineering approach, I organize a behavior AI has changed into a workflow, in four steps.

1 Find the behaviorWhich task have I started defaulting to asking AI first? How did I used to do it?

2 Observe the variablesWhat conditions does this task have, what needs judging, what are the exceptions? For an itinerary: what time it ends, which transport options, what to do if early or late.

3 Fix the processLay out the main route, the backups, and where I must confirm for myself. For variable information like train times, the final on-site check is the step I keep for myself.

4 Make it a promptOrganize it into a copy-paste question, so next time AI runs a first pass and I do the final check.

A ready-to-use prompt

If you want to try it right away, copy the block below to AI and replace the blanks with your own task.

I want to organize something I've already gotten used to handing to you into a reusable workflow.

Here's how I used to do this task: ______
The part I now hand to you is: ______
The conditions I need to judge along the way are: ______
The exceptions I might run into are: ______
The thing I still need to confirm myself is: ______

Please organize it into four things:
1. The complete workflow
2. The checkpoints I should oversee myself
3. Backup approaches for exceptions
4. A prompt I can copy and use directly next time

One last thingAI answering you once only saves the effort of that one time. It's when you settle it into a workflow that the task starts accumulating for you. Next time you notice you've gotten used to asking AI first for something, don't let it pass, that's a signal.

Turn a task into a system you can run again and again

Hand the building part to an agent workflow, and the auto-run part to program automation. The real point is whether you've kept a task as a process you can run again next time.

This is also what I keep doing: helping individuals and teams organize their work into processes AI can catch and run repeatedly.

Free Online Talks

Two free talks every month.

Join the LINE Community ↗