The fastest way to stall an automation effort is to start with the hardest, fuzziest process you have. The fastest way to build momentum is to start with the right small thing. Here is how to spot it.
The traits of a good first task
A strong first candidate is three things at once. Frequent, so the time saved adds up quickly. Well defined, so success is obvious and the agent has a clear target. And low blast radius, so a mistake is easy to catch and undo. Tasks that are rare, ambiguous, or irreversible make poor first choices, however tempting.
The usual winners
For most operations the early wins are the same. Chasing overdue invoices: frequent, defined, and the agent's reminders are easy to review. Inbox triage: it runs every morning and the summary is instantly useful. Stock reconciliation: repetitive and number-driven. New-hire onboarding: a checklist the agent can run end to end. Each of these is a place where a person currently spends time on coordination rather than judgement.
What to keep human, for now
Hold back the things that need taste or carry real risk until you have built trust: pricing decisions, sensitive customer conversations, anything legal or irreversible. You can still have the agent prepare the groundwork and route the decision to you. Human in the loop is the right setting for these, not human out of the picture.
How to expand
Once the first task runs cleanly, widen in two directions: deeper, by letting the agent handle more of the same area, and broader, by adding the next well-defined task. Before long the routine load is gone and what remains is the work that genuinely needs you. The guides walk through several of these step by step.