The older posts on this site say to start small and keep a person in the loop. This one is the list we use when we build. It comes from our own practice — the Responsible AI page and the automation lab — not from a client result. We have no verified client outcome to attach to it.
The line we use
We will automate a predictable step: saving a form, sending a reminder, sorting a routine request, or drafting something a person can check. We will not automate the decision in the middle when the situation is sensitive, unusual, or important. That line is written on the Responsible AI page /responsible-ai, and it is the same line we use in the work.
What the lab will not do
The automation lab asks a short set of questions and shows one starting point. Those answers stay in the browser. They are sent only if you choose Email me this plan, which opens the contact page /contact with the note attached. A quiz that filed a lead before you asked would break the promise on that page.
What the assistant will not do
Otto can explain what Zero1Two does. Otto is not allowed to invent a price, a timeline, or a guarantee. When the question needs a decision, Otto points to a person. We will not put a result figure in a field note unless we can stand behind it. These notes do not have one.
The work that still fits
The service that matches this note is Figure out where AI fits /services#ai-strategy. The plan starts with the easiest useful win and a check on what should stay with your team. The knowledge assistant example /case-studies/knowledge-assistant is an illustrative example of the same boundary: routine questions can use approved information, and anything unsure goes to a person.
If you want to talk
Tell us the task you are tempted to automate, and the part you do not want to hand over. That second part is usually the right place to start. Use the contact page /contact.
If a task needs judgment, a person keeps it. Software can carry the steps around that decision.