Every AI automation platform demos well. The question is what it does at 7 am on a Tuesday when nobody is watching and the request is one its maker never imagined. Before you connect your inbox, your calendar or your sheets, make the platform prove six things, in your account, on your data. Here is the list, and how Tensor Autonomous answers each.
1. It can state the job in one line
If the platform cannot say what it will do in a sentence, it will not do it reliably. Tensor's Library states every job in a line, for example: Logs each tenant's maintenance request with its urgency and drafts a reply that says what was logged. Jobs you describe yourself are held to the same standard: one sentence, in your words, that Tensor turns into a guide you can read.
2. It reads widely and writes narrowly
Ask what it will read and what it will change, separately. Reading the inbox a job lives in is fine; writing should be limited to the places the job names. A Tensor responsibility writes only to the accounts connected to it (the maintenance log, the leasing inbox's drafts), through Google's own permissions, which you can revoke in one click. Credentials are never shown to the agent as text.
3. It asks before it changes anything
Make it show you the exact change in your own account before the change happens: the event, the row, the email. 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. This is the default for every new job.
4. You can set limits, and it stops at the edge
Approving forever is not a plan. Ask whether routine changes can run within limits and what happens outside them. In Tensor you set, per responsibility, 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. At the edge it stops and asks instead of guessing. Sending email, payments and refunds, and moving or cancelling a calendar event always ask; limits can't cover them.
5. It leaves a receipt you can read
Not an event log. A page per run: what was asked, what it read, what it proposed, who approved it, what it wrote, and the read-back from the record showing the write matched. Tensor's receipt is written in plain words: which calendar or sheet it wrote to, when, and who approved it. It is where you go when a customer asks why.
6. There is a way back
Ask how a wrong change is undone and how a job that drifted is restored. In Tensor, next to each write, Undo removes what Tensor created after you confirm what will be removed; a responsibility can be paused, and its guide rolled back to an earlier version. The honest exception: sent email cannot be unsent, which is why every send waits for your OK; limits can't cover sending.
Two things not to be impressed by
- The number of integrations. Your work lives in a few places. Tensor works in Gmail, Google Sheets and Google Calendar today, with Microsoft 365 to come; what it does there is more useful than a logo wall.
- "Fully autonomous." Autonomy is a dial you turn as a routine earns it, not a feature you buy. A platform that starts autonomous has skipped step 3.
How to run the proof
Give the platform one real job for a week with approval on every change. Count how often it proposed the right change, how often it asked when it should have, and how long the receipts took to read. Tensor's live demo is that proof compressed into two minutes: a sample task in your words, the exact change shown, your approval, the write, the read-back.
Where to start
Related reading
- Workflow automation software: five questions before you buy
- AI agent governance for a small business
- The receipt: what Tensor records about every run
See it in a demo
The six proofs in one run.