Connecting your tools

Connecting Google Sheets to an AI agent

In most small businesses, the real management software is a spreadsheet. Site tracking, the members list, the delivery schedule, margin calculations: that is where it lives, and rarely anywhere else. The Google Sheets connector gives an agent the reading and updating of sheets designated one by one. This page states the two ways of connecting it — they are not equivalent — and why writing into a spreadsheet demands more restraint than anywhere else.

1. What the agent does in a sheet

The scope is deliberately narrow: read and update chosen spreadsheets. This is not access to your entire Google Drive, nor to your documents, nor to your email.

Read in order to decide. An agent that must chase unpaid invoices reads the due-date column rather than asking someone to read it out. An agent preparing an order checks the stock in the sheet where stock is kept.
Update what moves. Tick a processed row, enter a date, fill in a column left empty after a phone call.
Cross-check two sources. Compare what the sheet says and what another connected tool says — that is often where old errors surface.

The connection requires naming the Google account and each sheet concerned. There is no "all my sheets" mode: it would be quicker to set up, and far harder to defend six months later.

2. Two ways to connect, and which to choose

Google Sheets is one of the rare connectors to offer two paths. They have neither the same prerequisites nor the same comfort.

The official Google route. You connect your Google account, and the agent reaches the sheets that account can already open. It is clean, it is revocable in one click from your Google account, and the rights are exactly yours — never more. One caveat to know before committing: this Google connector is still in developer preview and must be enabled in a Google Cloud project. It is not an insurmountable operation, but neither is it "I click and it's connected". Better to know beforehand than to discover it halfway through.

The historical sharing route. Older, more rustic, and with no prerequisite at all: you share each sheet with nullbot's robot address, as you would share with a colleague, then you paste the sheet's link. Nothing to enable anywhere.

Our recommendation: if your organisation already has a Google Cloud project, take the official route. Otherwise, the sharing route gets you going today, and has a rarely mentioned advantage — the sharing is visible in the sheet itself, alongside the other people who have access. What is visible gets revoked; what is buried gets forgotten.

3. Why a write here deserves a review

A spreadsheet is not a database, and that is precisely where the risk lies. A database refuses a badly typed value; a spreadsheet accepts it. Three differences matter:

Writing overwrites. There is no easily recoverable "before" version for a given cell. Google Sheets history exists, but finding the right value of one cell among hundreds of revisions is not a common gesture.
One cell can feed a hundred. A value written into a column feeds formulas, charts and sometimes other sheets through cross-references. The consequence of a write goes far beyond the cell touched.
The structure is not guaranteed. A row inserted by a human in the wrong place, a column moved, and the reference point an agent was relying on has changed position without warning.

For these reasons, Google Sheets is classified at the high stakes level, and reviewing write actions before execution is the setting to keep. Our concrete advice: let an agent write freely in a sheet it feeds alone, and keep validation on any sheet where humans also work.

4. What it changes over time

The company spreadsheet has a well-known trajectory: impeccable at first, approximate after six months, abandoned after two years in favour of a new spreadsheet.

Week 1. The sheet stops being an evening manual entry. What must go in goes in at the moment the fact occurs, not when someone finds the time.
Month 3. Empty rows and duplicates become rare — not because someone did a clear-out on a Sunday, but because nothing accumulates any more.
Year 2. The sheet is still the original one. That is the real result: not having had to start over because the previous one was no longer reliable.

No figured gain is promised: it depends on the volume of your sheets and on the number of people who modify them.

5. Frequently asked questions

Does the agent see all of my Google Drive?

No. The permission covers spreadsheets, and the scope is named sheet by sheet at the moment of connection. Your documents, your presentations and your personal files are not concerned. With the sharing route, it is even clearer: only what you have explicitly shared exists for the agent.

What happens if someone modifies the sheet at the same time?

Google Sheets handles simultaneous editing as it does between humans. The point of attention is not the technical conflict but the reference point: if a column is moved while an agent is working, it may write in the right place according to the old structure. That is the main argument for keeping human validation active on shared sheets.

Do I really need a Google Cloud project?

For the official route, yes: Google's connector is in developer preview and is enabled in a Cloud project. For the sharing route, no — no prerequisite. Both lead to the same functional result; they differ in the setup and in where access is revoked.

Can you settle for reading only?

Yes, and it is often the right starting point. An agent that reads a sheet in order to answer accurately already brings the essential, with no risk at all to the data. Writing opens up afterwards, sheet by sheet, once the use is established.