01 · WHY AN EXTENSION

Add a small improvement
without changing how you work.

A Chrome extension can add a small feature to a web service you already use, without rebuilding the service itself.

And these days, AI makes software this small easier to build than it used to be.

  • 01 Build it small.
  • 02 A person checks it.
  • 03 Use it in the browser.

Those three, in that order, are the whole idea of this page. Reorder them and it becomes a different argument.

02 · BETWEEN YOU AND THE SERVICE

A Chrome extension sits between you and the web service.

A web service is built for its own purposes. The one extra step your team needs is not usually something you can ask the service to add for you.

An extension does not rebuild that screen. It lives in your browser and layers a small feature — a check, a bit of tidying — on top of the page you are already looking at.

You Typing, pasting, attaching, sending. The decision happens here.
Browser (where the extension runs) This is where the extension goes: on top of the page currently open.
Web service Unchanged. No rework, no configuration change, no migration.

The extension runs inside your browser. It is not reaching into the service.

The flip side is that its reach ends at the browser. Rewriting the service's own data, or replacing a core process, is not what this shape is for.

03 · AI × EXTENSION

Small software written with AI, and Chrome extensions, suit each other.

AI is bringing down the cost of writing software considerably.

That does not make "have the AI build the whole system" the best move. The wider the scope you hand over, the wider the surface you have to check afterwards.

Software that meets these conditions tends to work better with AI development:

  • It has a single purpose
  • The screen it acts on is well defined
  • Its inputs and outputs are small
  • The permissions it needs are explicit
  • A person can verify how it behaves

A Chrome extension is an easy shape in which to meet those conditions. One screen, one operation to assist. Implement that much and you have a working tool.

This is not the claim that AI lets anyone build an extension easily. Less writing leaves the harder work in place: deciding what to build, and whether it is fit to publish. What the shape gives you is a smaller thing to decide about.

04 · DEVELOPMENT LOOP

Generate with AI, check it, then use it.

In practice this is not a straight line but a short loop — and one step in it is deliberately not automated.

AI DOES THIS PART

A small problem in real work "On this screen, I check this every single time." Narrow it until it fits in one sentence.
AI agent / AI coding Not all at once: split the work and have it implemented in small pieces.
A small Chrome extension One purpose, one screen. At this point it only runs — nothing more.
Check the permissions, the code and the behaviour (a person, or a process)

What was generated does not go straight into use. It passes through here first.

  • manifest permissions
  • external requests
  • where data is stored
  • behaviour in a real browser

PEOPLE USE IT

Used in Chrome The usual browser, the usual screen. The steps people take stay much as they were.
Feedback The friction you only find by using it. Most of it never shows up in a code review.
Improve The area to change is small, so deciding whether to change it is light work too.
Once improved, you are back at defining the problem. How fast that circle turns is the advantage of working this way.

What matters in this figure is what it does not say. It does not say the AI builds all of it, automatically and safely. Between generating and using, a person looks.

Why this shape fits AI development

01

Responsibility stays small

You can decide that it assists one operation on one screen. That is a narrower target than making an AI understand an entire business system.

02

The browser is the runtime

For a feature that finishes inside the browser, you can get there without standing up a large server platform. Not every requirement works out that way.

03

Permissions are visible

A Chrome extension declares the permissions it needs in its manifest. Against code an AI wrote, that gives you one boundary at which to ask: what does this reach?

04

Behaviour is easy to check

Typing, clicking, copying, sending. The operations involved are reasonably clear, so you can try it yourself in a real browser.

05

Improvement stays small

Instead of finishing one enormous system, you get a short cycle: build, use, get annoyed, fix.

05 · NO BIG MIGRATION

It does not require a large system change.

Replacing a business system is large in cost, in time and in the number of people involved. For a problem the size of "I check this every time on this screen", a change of that scale is not realistic.

An extension leaves the existing screen where it is and adds on top of it. For the people using it, the steps stay much the same.

  • There are cases where you can start without any change on the service side
  • You can try it while keeping today's screens and steps
  • If it does not fit, remove it and you are back where you were

None of that means an extension will do. Processing that happens outside the browser, or requirements bound up with the consistency of the service's own data, are outside what an extension covers.

AI × CHROME EXTENSION × ENTERPRISE

Stack the three, and it becomes something you can keep doing.

Each stands on its own, but stacked in this order they turn into an approach that does not end at "built it".

LAYER 01 AI Builds small features on a short cycle. Split the spec, have it implemented, have a person check it.
LAYER 02 Chrome Extension Adds the feature to existing web work — delivered on top of the screen, without rebuilding the service.
LAYER 03 Enterprise Management Allows, distributes and controls it for the right people. Admins decide who gets it and what is permitted.

Combined, the three layers describe one approach: keep adding small improvements to daily work, without rebuilding the existing systems.

06 · PERSONAL, TEAM, COMPANY

Start alone, use it as a team, manage it as a company.

If most of the work happens in a browser, the extra step you want to remove is in there too. So the thing one person tried can spread to a team, and to an organisation if it needs to. The unit changes; the tool does not.

PERSONAL

Personal

Install it in your own browser to make your own work lighter. Nobody else has to agree, so you can try it today.

TEAM

Team

A few people who look at the same screen every day. Both the benefit and the side effects show up quickly, so you can decide quickly.

ENTERPRISE

Enterprise

Distribute it to a defined group and manage it by policy. What informs that decision is the permissions requested and a description of what it actually does.

07 · CHROME ENTERPRISE

In a company, admins can allow, distribute and control it.

Extensions are not necessarily something each person installs on their own. From the Google Admin console (Chrome Enterprise Core), an organisation can push policy to its managed Chrome browsers — so the question of using extensions at work can be discussed on that footing.

  • Choose the overall stance. In the Admin console you can allow everything and maintain a blocklist, block everything and maintain an allowlist, or run an allowlist that also accepts requests from users.
  • Name individual extensions. Policies such as ExtensionInstallAllowlist, ExtensionInstallBlocklist, ExtensionInstallForcelist and ExtensionSettings decide how each extension is treated.
  • Decide by permission. You can allow or refuse extensions on the basis of the permissions they request, and stop them running on particular sites.
  • Take requests. Extension requests from users can be allowed, blocked or force-installed by an admin.
  • Publish inside the organisation. There is a way to publish only to your own Google Workspace domain (your organisation's own Chrome Web Store).

How far any of this goes depends on the edition you have and on how it is configured. Check with your own administrators and with Google's documentation before deciding anything. The pages below are Google's.

08 · WE BUILD THIS WAY TOO

Legacy Tools is built this way as well.

What this site publishes is eight Chrome extensions. Each one is for a single moment — check before you send, replace before you paste — not a general-purpose business system.

They are built the way this page describes: write the problem as one sentence, have the AI implement it in separate pieces, load it into my own Chrome and use it, and check the permissions, the network calls and the storage before publishing.

What eight of them taught me is that the part AI can take on and the part a person has to decide are clearly separate. Deciding what to build, and whether it is fit to publish, takes longer than the building.

09 · PERMISSIONS AND LOCAL PROCESSING

"It is an extension, so it is safe" is not true. That is why you check.

An extension can ask for permission to read what is on a page, and to touch what you type. Being an extension is not in itself any guarantee of safety. You cannot leave that vague and still have a serious conversation about using one at work.

What you do get is the ability to check. The permissions it needs are declared in the manifest and shown at install time, so you can read what it reaches before you add it.

  • Which sites it runs on
  • What it stores, and where
  • Whether it sends anything outside
  • Whether each update asks for more than the last one

The eight tools here are designed to finish their work inside the browser and not send what you type anywhere. The permissions and the scope of processing are listed per product on the transparency page.

10 · ALONGSIDE EXISTING CONTROLS

It is not a replacement for DLP or audit logging.

A company already has machinery for how information is handled: DLP, logs, audit, access control. An extension does not replace any of it.

It sits somewhere else. Most controls act as data travels a path, or after it has travelled. An extension acts before that — in the seconds before a person presses send.

So they run side by side. Catching more at the front does not remove the need for what comes after; having what comes after does not remove the fact that a person still decides whether to send.

And the extension is itself subject to control: which extensions are allowed, and who gets them, is an admin decision, as in 07. "Unvetted extensions will pile up" and "improve the work with an extension" can be true under the same policy.

11 · START WITH ONE

Start with one piece of work.

You do not have to start from a company-wide plan. One screen, one operation, one team. At that size, you can undo it if it does not fit.

  1. 1

    Write the problem as one sentence

    On which screen, doing what, checking what. While that sentence stays vague, whatever gets built will be vague too.

  2. 2

    See whether something already does it

    An extension that does the same thing may already be published. If so, checking its permissions and behaviour and choosing it beats building your own.

  3. 3

    Build it small, then use it yourself

    One screen, one operation. Once it runs, use it in a real browser yourself. There is friction you will not find by reading the code.

  4. 4

    Check it, then widen it

    Check the permissions and the behaviour, then take it to the team. For an organisation, distribution and policy come in at that point. Reverse the order and it gets hard to walk back.

This is as far as the page goes. Not "AI makes it easy", not "extensions are safe". A Chrome extension is easy to design with a small responsibility and a small reach, and that is what fits a way of working where an AI implements and a person checks. That single point is the claim.