Many first attempts at AI automation fizzle for one reason: nobody decided what success would look like. Here is a first month with a shape: one job, a week of watching, a Friday habit, then a second job. It is how Tensor Autonomous is meant to be adopted, and it works for any tool that can be held to the same rules.
Week 1: pick one job and watch everything
Choose a job that arrives by email or lives in a sheet, ends in a small change, and that you do the same way every time. Tensor's Library has eight such jobs ready, each stated in a line; the gentlest first ones are Voicemail to callback list (turns each voicemail email into a row on your callback list, with the caller, number and reason) and Log maintenance requests (logs each tenant's maintenance request with its urgency and drafts a reply that says what was logged).
Run it with approval on every change. Tensor asks before it writes or sends: tensor shows you each change before it makes it, drafts included; nothing it drafts is sent until you send it. Read each one. You are not checking the agent's work so much as learning its habits.
Measure: how often the proposed change was right as proposed, how often you edited it, how often it stopped and asked.
Week 2: the Friday habit
Every run leaves a receipt: what was asked, what Tensor read, what it proposed, who approved it, what it wrote, and the read-back from the record. Spend five minutes on Friday reading the week's receipts. You will notice two things: the job is more regular than you thought, and the exceptions cluster.
Measure: minutes spent approving this week versus minutes the job used to take.
Week 3: set limits
For the routine you have watched run right, set limits: which kinds of change may run without asking (adding a row, creating a draft, adding an event), a daily cap, a days-and-hours window, and for each write an amount ceiling and a limit on recipients. Rows added to your log without asking, say, or bookings Tensor may make without asking on weekday mornings, up to three a day. Within the limits the work runs by itself; at the edge it stops and asks. Sending email, payments and refunds, and moving or cancelling a calendar event always ask; limits can't cover them.
Measure: how many runs asked you this week, and whether each ask was one you wanted.
Week 4: the second job
If the first job has been quiet for two weeks, hand over the second. Pick one that drafts to customers if the first did not (Reply to rental inquiries, Log new leads and draft a reply, Handle cleaning booking requests), and start it the same way: approval on everything, then limits.
Measure: whether you would notice if the first job stopped. If yes, it is doing real work.
What to leave out of the first month
Complaints, anything to staff, payments and contracts, and any job whose records are wrong today. Fix the sheet before you automate the sheet.
The way back, in case
Next to each write, Undo removes what Tensor created after you confirm what will be removed. A job can be paused while you read what happened, and rolled back to an earlier version of its guide. Sent email cannot be unsent, which is why every send waits for your OK; limits can't cover sending.
What it costs and needs
Plans start at $149 a month (Starter, Operator and Operations Team; Enterprise on request). Tensor works in Gmail, Google Sheets and Google Calendar today, nothing to install; Microsoft 365 comes later.
Where to start
- Your industry's starters: property management, agencies, home services and cleaning, event venues and vendors.
- Try it live: week one compressed into two minutes.
Related reading
- AI agents for small business: start with one job
- Workflow automation for small business: what to automate first
- Automated approval: what should run by itself, and what should wait for you
See it in a demo
One job, approval, receipt. The first month's shape in two minutes.