ChatGPT Can Find the Work It Should Be Doing for You
Use your calendar, sent email, and chats to find repeatable work, then automate the easiest task by steps, sources, and judgment calls.
- Ask AI to list recurring tasks from your data and cite evidence for each occurrence; don’t ask it what to automate.
- Choose a starter task with few steps, few information sources, and few judgment calls.
- Have AI interview you, test the written process together, then turn the refined workflow into a skill or project.
The walkthrough supplies reusable prompts and a concrete path from work evidence to a refined skill. · Best for: Professionals who want to identify and automate recurring work with AI.
The video presents a workflow for finding suitable AI tasks from actual work records, documenting how those tasks are done, and iteratively building a reusable AI skill or project.
Why “what should I automate?” is a poor starting point0:00
- That broad question can lead AI to suggest custom code or applications that are time-consuming and fragile.
- Start from your own recurring activities and consider built-in features such as skills and projects.
Break the role into activities, tasks, and steps1:46
- Map your role into activities, then tasks, then the steps, inputs, and outputs for each task.
- The video uses weekly reporting as an example of a task within a reporting activity.
Find recurring work in your records3:21
- Use calendar entries, sent email, and chats as evidence of recurring work.
- Connect ChatGPT or Claude to Gmail or Outlook, or export relevant chat histories for analysis.
Prompt AI to identify and rank tasks5:31
- Ask AI to describe each recurring task, count occurrences with dates, and cite the specific email or calendar event supporting each one.
- Limit results to 10 tasks, rank by frequency, and require AI to state when evidence is insufficient.
Choose a simple task and document its process7:01
- Compare candidate tasks by number of steps, sources, and judgment calls; start with the fewest of each.
- Have AI interview you one question at a time, up to 15 questions, and produce a one-page specification.
- Share examples of good outputs; the creator recommends dictating your answers.
Test, turn the workflow into a skill, and improve it9:55
- Try the task in a new chat using the specification, then ask AI to capture the working process as a skill or project prompt.
- Review results on future tasks and give corrective feedback to improve the AI’s judgment over time.
Step by step
- Review your calendar, sent email, and work chats; connect or export the relevant records.
- Ask AI to identify up to 10 recurring tasks, with occurrence dates and evidence citations.
- Count each task’s steps, sources, and judgment calls; select a task with relatively few of all three.
- Ask AI to interview you one question at a time, up to 15 questions, and produce a one-page process specification.
- Test the task with AI using the specification and examples of good outputs.
- Ask AI to convert the refined workflow into a skill or project prompt.
- Use it on new work, review its output, and give feedback when it makes mistakes.
Tools & resources mentioned
- ChatGPT and Claude data connectors — access Gmail or Outlook records for recurring-task analysis.
- Calendar, sent email, and Teams or AI chat histories — source data for identifying repeatable work.
- Chat export — provide prior ChatGPT or Claude conversations as evidence of recurring tasks.
- One-page process specification — capture steps, sources, judgment calls, and what a good output looks like.
- Skills and projects — built-in ways the video suggests packaging a tested workflow.
Notes are AI-generated from the video's transcript and may miss nuance.