Mathew PaquetTemplate · 5 min read
The Model Choice Sheet
Decide which AI model does which job, test it on your own file, and know what to check before anyone switches Claude on in Gemini Enterprise.
On October 8, Google said its newest model beats Anthropic's Claude on a coding test. The same day, it let its own customers switch Claude on. Two things launched:
Google Cloud's CEO, Thomas Kurian, put the idea in one line:
"Gemini is the agent, and the model underneath it is a separate choice."Think of a desk and a chair. The desk is the app: your files, your history, your instructions, who can see what. The chair is the model, the part that does the thinking. You keep the desk and change who sits in the chair, job by job. Microsoft put Claude next to OpenAI's models in Copilot, and Amazon sells OpenAI's open models. You don't need Gemini Enterprise for the rest of this sheet. It works with the AI tools your team already pays for.
For each job, ask: how hard is it? How often does it run? What data does it touch?
In Kurian's words: "The best model for the task is not always the largest one." Copilot already does the two-model version: one model drafts the research, a different one reviews it.
How hard is it? [ROUTINE / HAS TO BE GOOD]
How often does it run? [DAILY / WEEKLY / A FEW TIMES A YEAR]
What data does it touch? [PUBLIC / INTERNAL / CLIENT DATA]
Is this tool approved for that data? [YES / NO / ASK]
Model for this job: [MODEL]
Owner: [NAME]
Same real file, same instructions, two models. A demo file tells you nothing. Your messiest spreadsheet tells you a lot.
Instructions: [THE SAME INSTRUCTIONS FOR BOTH]
Model A: [ ] Model B: [ ]
Followed my instructions? [YES / NO] [YES / NO]
Got the numbers right? [YES / NO] [YES / NO]
Would I send it unrewritten? [YES / NO] [YES / NO]
What did it cost? [COST] [COST]
The best model will keep changing. Kurian says the leading model "changes every few months." Keep what makes the output yours on your own drive, not inside one vendor's app. Then switching models is a setting, not a migration.
Google's docs give the path: in the Google Cloud console, go to Gemini Enterprise, click Settings, open the AI developer tools tab, then Antigravity authorized models. Before you touch it:
1. Is our information sent to Anthropic to be processed?
2. How long is it kept?
3. We are on the [EDITION] edition. Does it include this setting?
Check the fine print. Claude runs in Google's global, U.S. and E.U. locations, not single-country regions. Google's docs say its coding tools don't support certifications like ISO 27001, an information security standard some clients require. For the Business edition, Google hasn't said.
Sources
Mathew's LibraryLibrary card
Keep reading free with your library card
One email opens this template and all 53 resources in Mathew's library. No account, no password.
Write down your three most common AI jobs and answer for each: how hard, how often, and what data it touches.
Saving works best in your phone's browser. A copy is also in your inbox.