Bottom line up front: doesn’t this look like a sensible way to integrate AI into Google Docs?

That’s interface0’s Google Docs integration. Just drop a comment tagging agent@interface0.com, same as you would a team member, and the agent can reply to your comment and even put suggestion-mode edits on the doc.

(If you want to try it out, sign up for the interface0 waitlist! You can then also email agent@interface0.com and get AI responses in email, WhatsApp its phone number to get responses there, and more…)

Mind-bogglingly, here is Google’s current Gemini integration into Google Docs:

Gemini can’t make suggestions or edit the doc at all, beyond telling you to manually click an “insert” button (and maybe also a “replace” button, although I’ve never gotten that one to appear). From the horse’s mouth:

How it is possible that Google — the creator of both Google Docs itself and of Gemini, one of the leading AI models, hasn’t yet integrated the two in any sort of sensible way?

The most obvious implementation is the native one: let the AI participate in comment threads and make suggestions on the doc like a team member could. And yet…

It turns out, by the way, that doing so (from outside Alphabet’s walls) is much harder than it seems.

Google’s Drive/Docs API, surprisingly, has no way to programmatically make suggestion-mode edits to a document. You can edit a document and can make comments on it programmatically, but can’t make suggestions. Feels like a big oversight…Plus they make it extremely difficult to identify which user is commenting / prompting, which creates a whole separate set of challenges.

So in order to get this working, I had to work out a more nuanced approach using browser automation.

Basically, when an interface0 user comments on a document and requests edits, a web browser instance is spun up in the cloud, agent@interface0.com signs into its Google Account, and then the browser programmatically navigates to the document in question and interacts with it to make the necessary changes generated by the AI pipeline on the backend. It’s imperfect and a little brittle, but has been working well enough for my purposes.

I hope Google adds suggestion-mode edits to the Docs API, but I’m not holding my breath…

Looking for more to read?

Want to hear about new essays? Subscribe to my roughly-monthly newsletter recapping my recent writing and things I'm enjoying:

And I'd love to hear from you directly: andy@andybromberg.com