All Things AI · Season 2 · The Models · Episode 6
Gemini’s advantage is not capability, it is location. If your documents, mail, and calendar already live in Google, the material is already where the model is, and not having to move anything is worth more in practice than most feature comparisons.
What it is
Google’s model, available as a standalone chat service and built into the Google applications a large number of agents already work in every day.
The standalone version is a capable general model comparable to the other major hosted services. The version that matters more is the one embedded where your material already is, because that changes the friction rather than the capability.
Why proximity beats features more often than people admit
This is the entire argument for it and it is worth taking seriously rather than treating as a consolation.
Moving material is the actual obstacle
Most AI tasks that never happen do not fail because the model could not do it. They fail because getting the material to the model was three steps and you had two minutes. A model already sitting where the document is removes the step that was stopping you.
It sees the context around a thing
A model with access to your mail and calendar can answer questions that require knowing what happened when, without you assembling that history first. That is a category of question that is impractical to ask anywhere else.
Nothing new to adopt
Working inside software you already use every day has a much higher chance of becoming a habit than opening a separate application. Most tools fail on adoption rather than on capability.
What to use it for
Anything where the material is already in Google and moving it is the friction.
- Questions across your own mail history, where assembling the thread by hand is the reason you never look.
- Working on documents in place rather than copying text out and pasting results back.
- Anything involving what was scheduled, when, and what happened around it.
- Quick drafting inside a document you are already editing, where opening another tool would break the task.
Where it is weaker
It is a hosted service, so the same rule applies as to every hosted service: material with a client’s details in it is better handled locally.
The convenience cuts both ways. A model with broad access to your mail and files is powerful and it is also a lot of access, and it is worth knowing what you have granted rather than accepting defaults.
And if your work does not live in Google, most of the argument for it disappears. This is a tool whose main advantage is contingent on where you already are, which is unusual and worth being honest about.
What to try in the first week
Two exercises, both aimed at the proximity advantage rather than at the model.
Ask it a question about your own mail history that you would never bother to answer by hand. When did this first come up. What did I say I would do about this and did I. What has this person asked me about across everything. Those questions have always been answerable in principle and never worth the twenty minutes.
Then work on a document without leaving it. Draft, revise, and restructure in place rather than moving text to another window and back. The point is to notice how much of your usual friction was transport rather than thinking, because that proportion is higher than most people expect.
Where this goes wrong
Three failures.
Granting broad access without looking at it. Convenience is the whole product here, and convenience is exactly the thing that gets people to click through permissions. Know what it can reach.
Assuming proximity means it is the right tool for everything. It is the right tool for material already in Google. For volume work, for privacy work, and for anything spanning a folder on your own disk, other things are better.
Putting client material into it because it was already open. Same rule as every hosted service, and the fact that the file is already sitting in your Google account does not change where the model runs.
How it sits next to the others
Against the other hosted services it is broadly comparable in capability and distinguished by where it sits. If your work is in Google, that is a real and daily advantage. If it is not, there is little reason to choose it.
Against local models it makes the usual trade: convenience and capability against privacy and permanence.
The practical version for most agents is that this is not a tool you adopt so much as one you start using because it is already there, and the honest question is whether your material lives where it is rather than which model is better.
The mail history question is the underrated one
Of everything in this episode, the ability to ask real questions of your own correspondence is the one most agents would benefit from and the one nobody uses.
Every agent has years of mail containing the actual history of every relationship they have. What was promised, what was asked, what the concern was two years ago, what happened the last time this came up. It is the most complete record of the business that exists anywhere.
And it is unusable, because finding anything in it means remembering enough to search for it, which means you can only find what you already knew was there. Search finds words. It does not answer questions.
A model reading across it answers questions instead, which turns an archive into something you consult before a meeting rather than something you occasionally dig through afterwards. That is a genuine change in what the material is for.
Keeping the boundary clear
A model with access to mail and documents is holding a great deal about other people, and that deserves a moment of thought rather than a default.
The useful framing is that this is fine for questions about your own work and your own history, and it is not the place for anything you would not be comfortable having processed by a service. Those are different categories and it is easy to stop noticing the difference once the tool is convenient.
In practice that means using it freely for what did I say I would do, when did this start, and what does this document need, and reaching for a local model when the question is about somebody’s file rather than about your own conduct of the work.
A note on tools you did not choose
There is something worth noticing about a model arriving inside software you already pay for, which is that you did not evaluate it and you will use it anyway.
That is not necessarily bad. Proximity genuinely is valuable and the barrier to trying it is zero. It does mean the decision about which model handles your work is being made by which software you happened to standardise on years ago, rather than by anything about the work.
Worth doing deliberately once. Ask what you are using it for, whether something else would be better for that, and whether the answer is only convenience. Convenience is a legitimate answer. It is a better answer when it was chosen.
What it costs
Included at a basic level with accounts many agents already have, with a paid tier for the more capable models. For anyone already paying for Google’s business tools, a meaningful part of this arrives without a separate decision, which is both its main advantage and the reason to think about it deliberately.
Questions agents actually type
What is Gemini best for?
Working with material that already lives in Google. Its advantage is proximity rather than capability, and not having to move documents or assemble mail history by hand removes the friction that stops most AI tasks from happening at all.
Is Gemini better than ChatGPT?
Broadly comparable in capability and distinguished by where it sits. If your documents, mail, and calendar are in Google, that daily proximity is a real advantage. If they are not, most of the argument for it disappears.
Is it safe to give Gemini access to my email?
It is a lot of access and worth understanding rather than accepting by default. Know what you have granted. For anything with client details in it, an open model running on your own computer is the better answer regardless of convenience.
Can Gemini answer questions about my past emails?
Yes, and this is the use most worth trying. Questions like when did this first come up, or what has this person asked me across everything, have always been answerable in principle and never worth the time to answer by hand.
Season 1 is Claude Code for Real Estate Agents, filmed and on its own hub. All three seasons: AI for real estate agents.