For ten years, a well-built website was an acceptable answer to the question "do I need an app?". This year, Apple and Google settled how the phone's assistant will act for us, and that answer no longer quite holds. The app is becoming an entry ticket again, and not for the reasons of 2008.
My colleague Dominique Siacci explained at the end of September why agents won't kill apps: on a phone, they can't act without them, because the app already carries the identity, the rights and the relationship an agent needs. I'd like to draw the consequence for the question we've been asked for fifteen years, and which is changing its answer for the third time.
Three answers to the same question
I've been building app platforms since 2011, and I've watched the answer to "do I need an app?" change twice.
The first answer was yes, no discussion. Between 2008 and 2014, being on the home screen meant existing. An icon on the phone was worth a shop window downtown. That rush is what gave birth to GoodBarber, and to dozens of tools like it.
Then the web took back control. Progressive Web Apps, Google pushing the browser as a platform, and a decade of "you don't need an app, a website is enough". The argument was solid. So solid that we added a PWA engine next to our native engines, in 2017, and we still sell it. One project description, three outputs: Swift, Kotlin, and the web. When I say the app had become optional, I say it as someone who built the option.
The third answer is taking shape this year.
What changed this year
At WWDC in June, Apple made App Intents the door through which Siri gets into third-party apps, and deprecated the old mechanism, SiriKit. iOS 27, released on September 14, launched a new Siri meant to act and search inside apps from a single sentence. It's early days. According to Apple, it's in beta and English first. It isn't available on iPhones in the European Union, and actions in third-party apps are coming "soon". Android is heading the same way with AppFunctions, shipped in Android 16 and still in beta with Gemini. Dominique covers the mechanism, so I won't go over it again.
The door itself isn't new: App Intents have existed since 2022. What changes this year is that Apple makes it the standard way in, and that the assistant will walk through it on its own, from a sentence.
For the question of this article, one thing matters: both mechanisms are native development kits, built into an app's binary, and a web page can't register with them. There is work on the web side, WebMCP, but it targets agents running in the browser, on an open page. The phone's assistant only looks at what's installed. I'm not saying anything about adoption, which nobody knows, or about when everyone will get access. I'm saying the entry ticket is defined, by both platforms, the same way.
Why this isn't 2008
It's tempting to read this as a step back.
In 2008, having an app was about capturing attention: an icon on the home screen, notifications. The app was a shop window. Tomorrow, having an app will be about being present in the assistant: when someone says "get me a seat for Saturday's concert", it's your app or someone else's that answers. The first was won with design and marketing. The second is won by describing your app, what it does as well as what it holds, precisely enough for a machine to use it.
There's a condition, and Dominique puts it himself on his list of what would prove him wrong: the assistant only calls what's installed. To be present in the assistant, you first have to be installed, and that's exactly the 2008 problem: the shop window is still the prerequisite.
This isn't the story I told in September either. There, I was talking about the cost of building: coding agents make it cheap to build the same app three times. Here, it's about distribution: who gets to be in the assistant. Two different mechanisms, pushing in the same direction.
So, do you need an app in 2026?
The question "do I need an app?" has had one answer per era. Yes, to be seen. No, a website is enough. And now: if you want to exist, one way or another, in the phone's assistant, yes.
First, to act. Booking a class, ordering, tracking a delivery, moving an appointment: these are exactly the sentences people will say to the assistant, and it can only carry them out in a native app that has declared those actions.
But also to be found. The assistant searches as much as it acts. When someone asks it to "find me a concert in Ajaccio next weekend", it will look in the content that installed apps have described to it: Apple explicitly plans for Siri to answer questions about an app's content. It's already what apps built on GoodBarber do when you ask Siri "what's new?", from iOS 17, without their owners having to touch anything. The rollout happens in stages, and my colleague João Aleixo explains how on our blog. A cultural events calendar or a local news outlet has as much to gain as a booking app.
And the assistant doesn't only live in the phone. In the car, few apps get a place on the CarPlay screen: Apple reserves it for a few categories, navigation or audio for instance, and rebuilding an interface for the dashboard rarely makes sense. But Siri is in the car, and an app that has declared its actions can answer there by voice.
The website keeps its role: it's what people find on Google, and it's what welcomes the visitor who comes once and won't install anything. We still sell PWAs, and many projects will have both: the website for Google, the app for the assistant.
It isn't urgent. If your users are in Europe, the new Siri hasn't reached them yet, and I don't know when it will. But the entry ticket is already set, and it's a native app.
The web had made the app optional. The assistant is making it mandatory for anyone who wants to be within earshot.
Mathieu Poli, Head of Frontend Engineering @ GoodBarber. I teach and write about frontend engineering, product design, and AI, and everything that happens when the three meet.
X: hellomathieup · LinkedIn: hellomathieup
Top comments (0)