Workflow Loom-to-Skill Pipeline
Turn recordings of real work into an AI-assisted workflow redesign.
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 90%
The Workflow Loom-to-Skill Pipeline converts demonstrations of real work into structured input for AI transformation. A team selects important workflows and records employees completing them while narrating decisions, tools, handoffs, exceptions, and pain points. The recordings are transcribed and combined with essential context: the team profile, current AI state, transformation goals, systems environment, constraints, and risks. An AI skill then analyzes the material through a framework such as Rapid Five to identify automation opportunities and propose before-and-after workflows. Employees review the output to correct missing context and reject unsafe assumptions. The result is a grounded redesign backlog based on observed practice rather than an abstract request to automate a department.
Origin
Extracted from Marketing Against The Grain as a practical way to simplify information gathering for an internal forward-deployed AI skill.
Core principles
- 01Observe real work instead of relying on generic process descriptions.
- 02Transcripts make tacit workflow details available for systematic analysis.
- 03AI recommendations need team context and transformation goals.
- 04Automation opportunities should be derived through a structured framework.
How to run it
- 1
Select Core Workflows
Choose five to ten workflows that materially affect the team's outcomes. Define where each workflow begins, ends, and hands off to another person or system.
Pro tip Prioritize frequent, costly, slow, or error-prone work.
Watch out Choosing an entire department as one workflow produces input too vague for useful redesign.
- 2
Record the Work
Have employees record themselves completing real examples while narrating decisions, tools, exceptions, and uncertainties. Capture the full path rather than only the cleanest portion.
Pro tip Ask employees to explain why they make each consequential choice.
Watch out Do not record secrets, personal data, or restricted material without appropriate authorization and controls.
- 3
Transcribe and Structure
Create transcripts and label the workflow, participant, systems, inputs, outputs, and major decision points. Preserve exceptions and workarounds that reveal hidden requirements.
Pro tip Link transcript sections to the relevant workflow stage.
Watch out Raw transcripts without context can cause the AI to misread banter, corrections, or unusual cases as standard procedure.
- 4
Add Transformation Context
Supply the team profile, current AI state, goals, systems environment, leadership context, constraints, and risks. Add skills inventory, benchmarks, and performance data when available.
Pro tip Clearly separate essential inputs from optional evidence so the exercise can begin without perfect documentation.
Watch out Omitting constraints and risk boundaries can produce unusable or unsafe recommendations.
- 5
Generate the Redesign
Give the transcripts and contextual material to an AI transformation skill. Ask it to identify where AI helps or hurts, then propose workflow-level before-and-after designs.
Pro tip Require the output to cite transcript evidence for each major recommendation.
Watch out Do not let the system equate technical automability with operational desirability.
- 6
Validate With Practitioners
Review each proposal with the employees who perform the work. Correct misunderstandings and move viable designs into measured real-world pilots.
Pro tip Use disagreement as evidence that more workflow observation may be needed.
Watch out Unvalidated AI analysis can automate a misunderstood process or remove necessary judgment.
In the wild
A sales representative records prospect research, meeting preparation, follow-up drafting, and CRM updates while narrating decisions and exceptions. The team transcribes the recordings and adds its tools, targets, risk constraints, and current AI usage. An AI transformation skill identifies repetitive research and data-entry steps, proposes a new workflow, and flags relationship-sensitive messages for human control.
→ The team receives a grounded automation backlog tied to real behavior rather than a generic list of sales use cases.
Common mistakes
Recording Only the Happy Path
Clean demonstrations omit the exceptions, uncertainty, and recovery work that often determine whether automation is safe.
Providing Transcripts Without Context
The AI cannot design a suitable operating model without goals, systems, constraints, and the team's current maturity.
Skipping Practitioner Review
Employees who perform the work must verify that the transcript interpretation and proposed redesign are accurate.
Is it for you?
Best for
It is best for knowledge-work teams whose processes are partly undocumented or depend on tacit individual practices.
Not ideal for
It is not ideal for workflows containing sensitive data that cannot be recorded, transcribed, or safely processed by the selected AI system.
From the transcript
“The skill requires essential team profile, core workflows, five to ten, current AI state and transformation goals. Like that's the bare minimum.”
“Important data and systems environment, leadership culture context, constraints, risks, and then helpful competitor benchmarks, individual skills inventory, and prior performance data.”
“And I think to your point, one of the suggestions would be have your team just do a bunch of looms of how they work…”
From the episode
This One Chart Exposes Why Most Companies Are Failing At AI