Spaces

Your first week of letting an AI agent answer user messages

This checklist walks a software company through the first week of an AI agent answering user messages: gather answers, set hand-over rules, test with real questions, then review every chat.

Runs in your browser. Nothing you enter is sent anywhere or saved.

Before you start

The first week with an AI agent is not about speed. It is about trust. You want to see your agent answer real user questions about trials, invites and billing the way you would, and you want to catch the odd wrong answer while the volume is still low.

Work through the steps in order. Most of them take under an hour. By the end you will have written answers, clear rules for passing chats to your team, and a habit of reading the conversations. Tick items as you go and keep the date next to each.

Tick each one as you go. 0 of 10 done.

[ ] Collect your twenty most common user questions from the last month.
[ ] Write a short answer for each, with the right help page link.
[ ] Write down your plans, trial length and refund policy in plain words.
[ ] List the situations that must go to a person and who takes each one.
[ ] Test the agent with ten messages from your own inbox history.
[ ] Try three tricky messages: an angry one, a vague one and a refund request.
[ ] Start with one inbox or account, not all of them.
[ ] Read every conversation at the end of each day this week.
[ ] Fix the answer, not just the single reply, when you find a mistake.
[ ] Tell your team which chats will arrive from the agent and how to reply.

What to look for on day five

By Friday you should have a feel for three things: which questions the agent handles cleanly, which ones it hands over too often, and which ones it should have handed over and did not. Write each down in one line.

Do not judge by volume alone. A quiet week with five excellent chats teaches more than a loud week you did not read. If something felt wrong, tighten the instructions and run the same test message again.

FAQ

Should we tell users they are talking to an AI agent?

Being open is usually the best choice. A short line such as 'I am the AI agent for our team, and a person can step in anytime' builds trust and sets honest expectations.

How many questions do we need written down before starting?

Twenty good ones are enough to begin. You will add more every week as you read real chats and notice what users ask that your notes do not yet cover.

What if the agent gets something wrong in week one?

That is normal and exactly why you read the chats. Fix the answer it used, send the user a short correction if needed, and test the same question again.

More free tools for software companies

Read the guide: Answer your users' messages with an AI agent.