MRT / Field guide 02 — Practical AI & systems
Your first useful automation.
An automation doesn’t need to be impressive to be valuable. It needs to remove a repeated bit of work without creating a bigger problem somewhere else. Start smaller than you think.
Find the task you keep doing twice.
For a few days, keep a list of repetitive work: copying information, renaming files, checking a status, or producing a recurring summary. Pick something frequent, predictable, and easy to undo. Avoid starting with payments, public posts, customer promises, or deleting data.
Write down the manual process.
Describe the trigger, inputs, steps, and finished output. Include the awkward cases: missing information, a duplicate file, or a source that isn’t available. If the process isn’t clear on paper, adding AI usually makes it harder to understand.
Use rules where rules are enough.
A formula, filter, or simple script may be all you need. AI can help with messy text or drafting, but it isn’t a substitute for an exact calculation or a permission check. Give it only the information it needs and keep sensitive material out of tools you haven’t evaluated.
Keep a human at the important step.
Let the first version prepare a draft or a preview. Review the result before anything is sent, published, charged, or changed. Keep an activity log, preserve the original inputs, and make it obvious when the system fails instead of silently accepting a partial result.
Measure the work it actually saves.
Try a small set of normal and awkward examples. Compare the result with your manual process, including the time you spend checking and fixing it. Keep the automation only if the total effort and risk are worthwhile. Expand one step at a time.