Three-Step Workflow Automation
Turn recurring manual work into an AI-operated system
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 98%
The method begins with observation rather than tooling: review the previous week and identify work that consumed substantial time or felt needlessly manual. Choose one recurring task, then document every action exactly as it currently happens, including research, transfers between applications, formatting, review, and publishing. Feed that sequence to Codex or Claude Code and collaborate with the agent to replace suitable actions with skills, APIs, browser use, or computer use. The goal is not immediate, unattended automation; it is a reusable system that removes mechanical work while preserving human review where taste or judgment matters. Test the system on a real instance, identify breakdowns, and revise it before attempting to automate additional workflows.
Origin
Peter Yang described this three-step method on Marketing Against The Grain while demonstrating how he converted his manually operated podcast process into a chain of reusable Codex skills.
Core principles
- 01Automate work that recurs or creates persistent friction
- 02Document the real process before redesigning it
- 03Treat AI as a system builder rather than a chat interface
- 04Start with one workflow whose removal would create meaningful time
How to run it
- 1
Review Recent Work
Look back over the previous week and identify tasks that consumed the most time or created the most frustration. Favor work that recurs frequently enough to justify an initial setup investment.
Pro tip Review completed work and calendar activity rather than relying only on memory.
Watch out Do not automate a task merely because it is easy to describe; prioritize meaningful time savings.
- 2
Map the Manual Workflow
Write every action in the order you perform it, including application changes, research, formatting, approvals, and publishing. The list becomes the agent's working specification.
Pro tip Describe the current process before trying to improve it.
Watch out Omitted steps become hidden assumptions that can cause the automation to fail.
- 3
Build the System With AI
Give the complete step list to Codex or Claude Code and ask it to streamline the process. Let it propose skills, scripts, APIs, browser actions, and integrations appropriate to each step.
Pro tip Ask the agent questions in ordinary language when you do not know which technical mechanism to use.
Watch out Do not assume the agent's first implementation is production-ready.
- 4
Run a Real Test
Execute the workflow on an actual task and inspect every output and handoff. Record any missing context, weak output, or unnecessary manual action.
Pro tip Use a representative case rather than a simplified demo.
Watch out A system that works only on an ideal example is not yet reusable.
- 5
Refine and Reuse
Update the workflow from test feedback, then reuse it whenever the task recurs. Expand to another workflow only after the first one produces dependable value.
Pro tip Preserve human review at steps involving taste, judgment, or consequential decisions.
Watch out Trying to automate an entire personal operating system at once can create complexity without savings.
In the wild
Peter mapped guest research, interview-guide preparation, Riverside recording, editor handoff, transcript retrieval, newsletter creation, social posts, thumbnail copy, titles, show notes, and scheduling. He then created multiple Codex skills and chained them into a podcast workflow.
→ The system reduced copying, formatting, and asset-production work while leaving Peter in control of creative judgment.
A marketer reviews the previous week, notices that weekly campaign reporting requires collecting metrics from several websites, and lists each source, calculation, formatting step, and delivery action. Codex turns the sequence into a reusable reporting workflow with a final human check.
→ The marketer recovers time each week without sacrificing responsibility for the report.
Common mistakes
Automating Before Mapping
Starting with tools before documenting the real process causes the system to miss dependencies and hidden manual steps.
Choosing a Low-Value Task
Automating an infrequent or trivial task can consume more setup time than it ever returns.
Expecting One-Shot Perfection
Skills and integrations require testing, feedback, and human review before they become dependable.
Is it for you?
Best for
It is best for knowledge workers who repeat multi-step digital tasks at least weekly.
Not ideal for
It is not ideal for rare tasks, unstable processes, or work requiring fully human judgment at every step.
From the transcript
“number one, just reflect on your past week.”
“Step two is just like literally lists out every single step of that manual workflow.”
“And then use Codex or Claude code to turn into a system.”
From the episode
Automate Boring Tasks With Codex & Claude Code in X Minutes