AI-First Task Transformation
Turn recorded work into a scoped, costed AI-assisted workflow
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 97%
The method begins with evidence rather than abstract automation ideas: record a person completing real work while narrating decisions, tools, and friction. An AI transformation skill analyzes that record, identifies whether the opportunity concerns one task, an individual, a team, or a department, and adjusts the depth of its output accordingly. It reconstructs the present workflow, estimates its duration and pain points, and proposes an AI-assisted alternative. The resulting comparison covers steps, tools, time savings, quality improvements, costs, and implementation actions. For simple work, the output may recommend a human using an AI assistant; complex work may require agents, reusable skills, infrastructure, and shared processes. The report then becomes a structured input for building and piloting the new workflow.
Origin
Extracted from Marketing Against the Grain, where the hosts demonstrate an AI transformation skill using narrated recordings of real work.
Core principles
- 01Observe real work before redesigning it
- 02Match the analysis depth to the scope of the work
- 03Compare the current and transformed workflows explicitly
- 04Estimate both efficiency gains and quality improvements
- 05Convert recommendations into small implementation steps
How to run it
- 1
Capture the Current Work
Record the work from start to finish and narrate the actions, decisions, tools, and difficulties. Preserve the actual duration whenever possible.
Pro tip Use an ordinary screen recording rather than preparing an idealized demonstration.
Watch out A short description may force the system to estimate timing instead of measuring it.
- 2
Determine the Transformation Scope
Classify the input as a task, individual, team, or department transformation. Use that classification to determine the necessary depth of analysis.
Pro tip Keep a simple task report short while allowing complex team transformations to receive more detail.
Watch out Applying department-level analysis to a small task can make the output unnecessarily overwhelming.
- 3
Reconstruct the Existing Workflow
List the current tools, steps, handoffs, duration, evaluation criteria, and pain points. Confirm that participants agree this is how the work actually happens.
Pro tip Treat disagreements about the current process as useful operational findings.
Watch out Do not automate an assumed process that differs from real practice.
- 4
Design the AI-Assisted Workflow
Identify which steps AI can accelerate, improve, or automate. Decide whether the work needs an assistant, reusable skills, agents, or supporting infrastructure.
Pro tip Separate simple assistant use from genuinely agentic automation.
Watch out Avoid recommending complex infrastructure when a single assistant can handle the task.
- 5
Compare Before and After
Estimate the transformed workflow's time, cost, and quality against the current baseline. Distinguish pure efficiency gains from genuine quality improvements.
Pro tip Use measured recording duration as the baseline when available.
Watch out Label inferred durations as estimates rather than presenting them as measured facts.
- 6
Run a Small Pilot
Turn the recommendations into concrete next steps and test the new workflow for a week. Use the pilot results to refine prompts, skills, and automation.
Pro tip Begin with one recurring workflow that creates visible value.
Watch out A detailed report has little value if nobody converts it into changed behavior.
In the wild
A CMO records a two-minute explanation of reviewing slides, synthesizing comments, and drafting an email. The transformation analysis maps the manual process, identifies Claude as the only necessary tool, and proposes a shorter workflow in which AI evaluates the material, collates feedback, and drafts the message.
→ The projected duration falls from 25–45 minutes to 8–15 minutes, with an estimated weekly recovery of one to two and a half hours.
A three-hour Loom recording shows an analyst working in Claude Code. The analysis reconstructs the setup and data-analysis process, then recommends a bootstrap script that creates directories, imports documents, and converts them to Markdown before the substantive analysis begins.
→ Repeated preparation work becomes automated infrastructure rather than a manual prelude to every analysis.
Common mistakes
Transforming an Imaginary Workflow
Designing from job descriptions or assumptions conceals the actual tools, exceptions, and informal decisions used during real work.
Producing Too Much Detail
A long transformation report for a small task can overwhelm the user and prevent implementation. The output should scale with the scope of the work.
Stopping at the Report
The analysis is an input for prompts, skills, agents, and changed habits, not the final outcome itself.
Is it for you?
Best for
Individuals and teams with repeatable knowledge work that involves multiple manual steps or tools.
Not ideal for
One-off activities whose process, inputs, and desired outcome cannot be observed or described.
From the transcript
“you can record yourself just doing work and then narrating your work and you can give that to an agent, and that agent will build…”
“I basically have it, is this a task transformation, a deeper like individual level transformation, a team level transformation, or a department level transformation?”
“No matter what you put through the skill, you're going to get a report like this, which is basically what the heck am I trying…”
From the episode
Marketing is Already Dead, You Just Don't Know It