By the end of this chapter you can take a task you do over and over and see it as a repeatable pipeline, which is where real compounding comes from, instead of doing it from scratch each time.
You can build, judge, and produce real things. This chapter is about the mindset that multiplies all of it. Most people think in tasks: this email, this report, this post. The people who get the most out of AI think in pipelines: the repeatable flow that turns out emails, reports, posts, again and again, without starting over each time.
The scarce skill is not prompting and it is not coding. It is workflow thinking. A task is something you do once and forget. A pipeline is something you set up once and reuse forever. The same hour of effort produces a one-off result or a machine that produces results. Learning to see the pipeline inside the task is the habit that compounds the most in this whole course.
AVATAR OPENER · ~90s
Watch: turning a task you redo every week into a flow you run once
HeyGen avatar · generated, consistent presenter
Here is the tell that you are stuck in task thinking: you notice you are doing the same kind of thing again, and you do it from scratch again anyway. Every Monday report. Every client onboarding. Every video description. Each one feels like new work. It is not. It is the same work with different inputs, which is the exact definition of something you can turn into a pipeline.
Thinking in tasks
Each job is done from scratch. Same kind of work, every time, with no memory that you did it before.
Result: you work hard forever and never get ahead of it.
VS
Thinking in pipelines
You set up the flow once: the steps, the rules, the shape. Then you just feed it new inputs.
Result: set up once, run forever. The work compounds.
The scarce skill is not prompting or coding. It is workflow thinking.
A task you do once. A pipeline you set up once and run forever. The same hour buys a result or a machine that makes results.
TURNING A TASK INTO A FLOW
You already have the pieces for this. A pipeline is just your earlier skills, arranged. The move is to design it before you build it, which is the same discipline as Chapter 13, applied to a repeating job.
Spot the repeat
Notice the job you keep redoing from scratch.
Name the steps
Write the steps you actually follow each time.
Make it a skill
Teach the flow once, as a skill, so it triggers on demand.
Feed it inputs
Run it on new inputs forever. The thinking is done.
Before you build any pipeline, one habit the best builders never skip: stress-test the plan first. Have the model interview you about the workflow, hard, before you commit to it. This is the grill-me move from Chapter 9, turned on your own process. It surfaces the step you do without realising, the edge case you forgot, the thing that would have broken the flow on week three.
SEE IT
Here is the prompt that designs a pipeline with you. Notice you are not asking it to build anything yet. You are asking it to find the holes in your process while changing them is still free.
The spec-it-out prompt (stress-test the workflow)
I do this every week and I want to turn it into a repeatable flow:
[describe the recurring job].
Before we build it, grill me. Ask me one question at a time about the
exact steps I follow, the inputs each time, what good looks like, where
it usually goes wrong, and what would break the flow.
Keep asking until you could run this without me. Then propose the
pipeline as clear steps. Do not build it until I approve the steps.
What you get back is your own process, written down clearly for the first time, with the gaps exposed. Half the value is that most people have never actually articulated the steps they take. Once it is on paper and stress-tested, turning it into a skill is the easy part. The thinking was the work, and the thinking is done once.
NOW YOU TRY · CREATE
Turn one repeating task into a pipeline
Find one job you genuinely do over and over from scratch. Use the spec-it-out prompt to have Claude grill you about your real steps until it could run the flow without you. Approve the pipeline, then turn it into a skill (Chapter 8) so you can trigger it. Run it once on a fresh input and confirm it produced the result without you re-explaining the process.
Right if running your pipeline on a fresh input produces the finished job in your steps, and the grilling surfaced at least one step you do automatically without having written it down.
Show the worked solution
The reveal in this drill is almost always a hidden step. Take a freelancer who writes a weekly client report. They think the process is "summarise what I did." But when the model grills them, it comes out in layers: first they check the project notes, then they pull the three things that actually moved, then they flag anything at risk honestly, then they translate it out of jargon for a non-technical client, then they end with next week. Five steps, and they had only ever named one. That is the value of the stress-test: the workflow you run on autopilot is more detailed than you realise, and a pipeline that misses the hidden steps produces thin work. Once all five are on paper and approved, you turn them into a skill that triggers on "draft this week’s client report", and from then on you feed it the week’s notes and it runs the full flow. You went from doing the report to operating the report. Do that to five recurring jobs and you have quietly built yourself a small operation, run by you, one input at a time.
WATCH FOR
✗You keep redoing the same job from scratch. That is the signal, not the cost of doing business. Same work, new inputs, means make a pipeline.
✗You build the flow before stress-testing it. Grill the plan first. The step you forgot is cheaper to find on paper than on week three.
✗You never write your actual steps down. You run more steps than you think. Get the hidden ones out before you automate.
✗You think leverage is a better prompt. Leverage is the pipeline. One good flow beats a hundred good one-off prompts.
WHAT YOU LEARNED
The takeaways
The scarce skill is workflow thinking. Most people think in tasks; leverage comes from thinking in pipelines.
A task is done once. A pipeline is set up once and run forever, so the same effort builds a machine instead of a result.
Turn a task into a flow: spot the repeat, name the real steps, make it a skill, then feed it new inputs.
Stress-test the workflow before you build it. Grilling surfaces the hidden steps you run without realising.
Your project · step nineteen
Look at your thread project and find the one job inside it you will do repeatedly. Turn that job into a pipeline: stress-test it, write the steps, make it a skill. Your project stops being a thing you operate by hand and starts being a flow you run. That is the operator shift. The last two chapters are about doing it honestly and sustainably.
Anyone can do a task with AI. The operator notices the task is a pattern, and builds the pipeline once. That noticing is the whole skill, and now you have it.