Small helpers in the browser have so far been a marginal topic: anyone who wanted an extension needed either development skills or a suitable product from a store.With the upcoming system versions macOS, iOS and iPadOS 27, this starting point is changing.Safari is to make it possible to create browser extensions via an AI prompt – that is, via a description in plain language instead of written code.

This brings an approach to the operating system level that is discussed in developer circles as vibe coding: you describe the desired function, a language model generates the code, and the result is judged primarily on its behaviour, not on the implementation.What is new is less the technology than the place: if such a function sits in the standard browser of a widely used operating system, the barrier to entry for many users drops drastically.

What this means in practice

Browser extensions are typically small programs that modify or supplement web pages in the browser – for example by hiding elements, highlighting content, shortening recurring click paths or preparing data for further processing.It is precisely in this area that small needs constantly arise in companies that never make it into a project budget.

Examples from everyday work are quickly found: a field in an internal web tool that should always be pre-filled with the same value.An overview page where three of twelve columns are in the way.An approval process in which you regularly jump back and forth between two views.Such requirements are too small for a ticket with the service provider, but in total they cost working time.

If such adjustments can in future be created through a description, the question shifts: no longer Can we build this? is in the foreground, but Do we want to run this permanently?

Where the limits begin

Prompt-generated code has known weaknesses that do not disappear even with a convenient interface.Three points are relevant for operational use.

First, quality. A result that works in the first test is not automatically robust.Edge cases, unusual page structures or changes to the target application lead to an extension silently no longer taking effect.Especially with tools that modify or pre-fill data, a silent error is more unpleasant than a visible crash.

Second, maintenance. Anyone who creates an extension via prompt still owns code at the end – only nobody in the team has really thought it through.If the colleague who created the tool leaves the company, an artefact remains without documentation and without a clear owner.With several such building blocks, a shadow IT emerges that does not appear in any inventory.

Third, security. Browser extensions naturally work close to the content that is displayed in the browser.This can also be customer data, personnel data or internal documents.Which permissions an extension requests, where it may send data and which external dependencies it brings along cannot be inferred from a prompt result by itself.A deliberate review is needed here before a tool is distributed beyond one's own computer.

A sensible approach for companies

The answer to these risks is not a ban.Anyone who categorically prohibits such functions usually only achieves that they are used invisibly.What is more sensible is a clear classification by purpose.

  • Non-critical: purely visual adjustments on one's own device that store nothing, send nothing and change nothing – such as hiding disruptive elements or highlighting information.
  • Requiring review: everything that accesses internal systems, reads out data, processes it further or transmits it to third parties.This includes a look at the generated code and the requested permissions.
  • Not suitable: functions that are to become a fixed component of a business process.What many people use daily needs versioning, tests and a responsible person – that is, the normal development path.

A simple registration requirement is also helpful: anyone who uses such a tool in a work context enters it in a list – with purpose, creator and affected systems.The effort is small, the effect for traceability and audits considerable.

The real gain lies in prototyping

The prompt-to-code function is most valuable where it makes ideas quickly testable.Instead of discussing a possible improvement in a workshop, a working variant can be tried out within a short time.If it proves itself in everyday use, the requirement is subsequently described much more precisely than any concept paper could achieve.

In this role, the approach does not replace development but improves its preliminary stage.The result of the prompt is then not the product but the specification – a runnable draft that shows what is really needed.

The signal emanating from integration into an operating system is also remarkable.Functions that previously required specialised tools are moving closer to the standard equipment.

What applies to browser extensions today is likely to apply to other small automations in the medium term.

Classification

For web and software projects, this shifts the boundary between requirements and implementation: business departments can make initial versions visible themselves, while development teams are needed more for architecture, security and long-term operation.Anyone who wants to benefit from this should decide early on which tools may remain in the personal environment and from what point a project belongs in regular development.Without this demarcation, many small solutions arise that are useful individually and remain difficult to justify in sum.

Sources