Start with repeated steps

Choose one daily task with clear inputs and results. Making it run reliably matters more than adding many capabilities at once. This draft reflects my project direction and will be reviewed before publication.

I focus on defining who does the work, where inputs come from, who receives the result, and which decisions cannot be automated.
‍

‍
Make tools serve the task

Tools can reduce repetitive work, but exceptions and business decisions still need people. Making that boundary clear prevents a demo from being mistaken for a complete delivery.

I want development and operations in the same room, observing how tools change the work instead of merely counting generated pieces.

‍

Check these five things first

Before automating a task, write down these questions:

• Where do assets and data come from?
• Which steps repeat every day?
• Which decisions need a person?
• Where to resume after failure
• How will the result be checked and handed over?

‍

My process
‍

Define the input
Confirm assets, format, usage rights, and quality requirements early to avoid repeated requests later.

Narrow the use case
Test one clearly valuable workflow before trying to cover every department and task.

Keep human review
Keep budget approval, key actions, and final quality checks as explicit human handoff points.

Record failures
Record execution status and errors. Distinguish resuming a task from starting again to avoid duplicate work.

Back to operations
Review the workflow with the people using it and see whether the saved steps truly help.
‍

Continue testing

I will continue testing these methods in XPLA, SoraTK, and real collaborations. I do not turn unmeasured efficiency into numbers or count unfinished directions as delivered outcomes.