Spiral Example-to-Workflow System
Encode proven examples into reusable AI transformation workflows
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 7
- Confidence
- 98%
Spiral turns a recurring content transformation into a reusable team workflow. The builder selects a clearly bounded task, such as converting an article into headlines, and supplies approved examples containing both the original input and the successful output. The system extracts patterns from those pairs and uses them to construct a prompt specialized for that conversion. Team members can then submit new material and receive a draft informed by the organization's established taste and voice. The output is not treated as finished work; it starts experienced writers closer to a viable result so they can spend more time polishing, exercising judgment, or creating something new. Because the examples and inferred patterns remain legible and editable, the workflow can evolve as editorial standards change.
Origin
Extracted from Marketing Against The Grain. Dan Shipper described Spiral, an internal Every tool built to automate recurring editorial transformations from approved examples.
Core principles
- 01Start from approved examples rather than abstract instructions alone
- 02Use AI for repeatable input-to-output transformations
- 03Make organizational taste visible and editable
- 04Let specialists polish strong drafts instead of repeatedly facing blank pages
- 05Keep humans responsible for the final quality threshold
How to run it
- 1
Define one conversion
Select a recurring task with a recognizable input and output, such as an article becoming a headline or a podcast transcript becoming show notes.
Pro tip Begin with a narrow transformation that happens frequently enough to justify standardization.
Watch out Combining several unrelated transformations will make quality harder to diagnose.
- 2
Collect approved pairs
Gather examples showing both the source material and the output that the team considers good. Favor examples that represent the current voice and standards.
Pro tip Use diverse examples within the same task so the workflow learns principles rather than one template.
Watch out Weak or outdated examples will encode weak or outdated taste.
- 3
Extract the patterns
Have the system identify the recurring structural, tonal, and editorial relationships between each input and output.
Pro tip Review the extracted patterns and rewrite any that are vague or inaccurate.
Watch out Do not confuse accidental similarities in the examples with deliberate editorial rules.
- 4
Build the reusable workflow
Turn the accepted patterns and examples into a prompt or internal tool that consistently accepts the chosen input format.
Pro tip Expose the examples and rules so editors can inspect and change them.
- 5
Test unseen material
Run new inputs through the workflow and compare its drafts with what experienced team members would produce.
Pro tip Track recurring failure modes and address them by improving the examples or rules.
Watch out Performance on training examples does not prove that the workflow generalizes.
- 6
Start humans at 70 percent
Give the generated draft to a skilled team member for judgment, correction, and final polish rather than publishing it automatically.
Pro tip Redirect saved time toward higher-value creative work and quality improvements.
Watch out A plausible draft can still contain factual, tonal, or contextual errors.
- 7
Maintain the taste library
Replace examples and revise patterns as the organization's voice, audience, or quality standards evolve.
Pro tip Review the library after major editorial changes or repeated output failures.
Watch out An unattended workflow will gradually diverge from current expectations.
In the wild
An editor supplies several strong Every articles paired with their published headlines. Spiral extracts the recurring headline patterns and creates a shared transformation. When a writer pastes in a new article, the workflow returns several candidates that already reflect much of the publication's established approach.
→ Writers begin with relevant headline options and reserve more time for selection and refinement.
A media team pairs previous podcast transcripts with approved promotional posts. It builds a workflow that identifies each episode's central claim and drafts several posts in the team's voice. A producer verifies the claims, removes generic wording, and publishes the best revision.
→ A repetitive repurposing task becomes faster without removing editorial review.
Common mistakes
Training on unapproved examples
The workflow reproduces patterns from what it receives, so mediocre examples institutionalize mediocre output.
Expecting a finished product
Shipper described the model as getting much of the way there, not replacing the writer's final judgment and polish.
Hiding the encoded taste
If people cannot inspect or update the examples and patterns, the workflow becomes an opaque standard that is difficult to improve.
Is it for you?
Best for
Teams repeatedly converting one known content format into another, such as articles into headlines or podcasts into show notes.
Not ideal for
Novel creative work whose desired output cannot be demonstrated with credible examples.
From the transcript
“if there's any kind of like task like that where you're taking an input and turning into an output reliably it helps you build a…”
“it will pull out the patterns in those examples and then it will create a spiral that anyone can use”
“for those writers it sort of starts them at like 70%”
From the episode
How I Use ChatGPT To Grow My 54,000+ Subscriber Newsletter