The Build or Buy Card – Mathew's Library

01 What a private AI app can and can't do

Google AI Studio is Google's tool for building with its Gemini models. Since May it could turn a description into a native Android app you try on a phone that runs in your browser. On October 13, Google removed the last step: a normal Google account is enough to put that app on your own Android phone through Google Play, privately. It starts in the U.S., for people 18 and over, and for now the app runs on your device only.

Before anyone on your team builds a work tool this way, know the limits:

It can use the phone's camera, location and Bluetooth, and call Gemini inside the app. It can't keep one shared copy for a team. Google's documentation says these Android apps are client-side only, with no server behind them. It can't connect to your company's Google Docs or Sheets, or hold a password or key safely. It can't be shared yet. Google says sharing with "a close circle of friends and family" is coming "in the coming months."

Private on the store isn't private in the app. If the app asks Gemini something, what you typed goes to Gemini. And Google hasn't said whether a company-managed Google account can use this at all.

02 The three questions

Building the first version got easy. Owning it didn't. Run every tool your team wants through these three before anyone builds it:

Who else uses it? Passes if one person uses it. Fails the moment a team relies on it: it needs one shared version everyone trusts, and this app has nowhere to keep one.

Whose data goes in? Passes with your own numbers (measurements, material prices). Fails with a customer's names, addresses or phone numbers: that's customer data in an app nobody reviewed, on a phone that can be lost, outside anything your business backs up.

Who fixes it, and what happens when they leave? Software you buy comes with a team whose job is to keep it working. Software you build comes with you. If the builder leaves, the tool leaves on their phone.

Why question two matters now. In a 2026 survey PagerDuty commissioned, 34% of the office workers surveyed had put customer data into public AI tools; at companies under 1,500 people it was 40%.

03 Build, buy, or build it as the brief

If the tool... Then Example

Passes all three: one person, nothing sensitive, answers you can check Build it A quote calculator for your own jobs

Fails question one or two: more than one person, or customer data Buy it An intake form, a customer list, anything near payments

Needs to be shared and safe, but nobody can describe it yet Build it as the brief A rough version with made-up data, handed to a vendor or developer

When you buy, what you're paying for isn't the features. It's somebody else keeping the tool shared, safe and working.

When you build, or build the brief, start with one sentence written like a spec. Paste this into Google AI Studio and fill it in:

The spec sentence

Make me a [NAME OF TOOL] for [WHO USES IT].
What goes in: [INPUT 1], [INPUT 2], [INPUT 3].
What comes out: [THE ONE RESULT, e.g. one price].
A wrong answer looks like: [e.g. a price below my cost, or a blank field accepted].

For the brief, use made-up data only, keep going until it does exactly what you want, then hand it over and say: this, but shared, and safe.

04 Test it before you rely on it

AI-written code isn't automatically sound. In a Veracode test, when AI models had a safe and an unsafe way to write the same code, they picked the unsafe one 45% of the time. So check the tool the way TechRepublic suggests:

Run numbers you already know. Put three old jobs through it and compare the result to what you actually quoted.

Close the app and reopen it. Check that what you entered is still there.

Try a wrong answer on purpose. Enter the case from your spec sentence and see whether the app catches it.

Notice the Monday. An app for one person is the right size for a lot of tools. The day someone else asks for it, run question one again.

Sources

Google: Google Play updates (October 2026) Google AI for Developers: Build Android apps in Google AI Studio Android Developers Blog: Build native Android apps in Google AI Studio TechCrunch: Google Play now lets anyone build private, vibe-coded apps that run on Android TechRepublic: Google Makes Personal AI-Built Android Apps Easier to Run PagerDuty: Shadow AI workplace survey 2026 Veracode: AI-generated code security research

Mathew's LibraryLibrary card

Free access

Keep reading free with your library card

One email opens this checklist and all 3 resources in Mathew's library. No account, no password.