Swift Stories

With LLM-based technologies becoming more widespread, and models getting more sophisticated and cheaper, people are getting used to a new type of interface that's not made by pixel but by language. It's still early, but there's a high possibility that many of the interactions that we know today in the context of apps (tab bars, buttons, navigation, etc.) will be replaced by natural language interactions. Can you imagine booking a flight or ordering a pizza with just a few words? To enable that future, it's necessary a standard interface that eliminates the MxN problem that can emerge, where there are multiple interfaces to communicate between a chat application like Claude or OpenAI, and the app itself. Anthropic has realized the need for it, and has gone ahead proposing a protocol, MCP for this purpose. An app can share resources, tools, and prompts (among others) that the LLM can interface with. @mattt (NSHipster) talked about it recently and shared his new project iMCP where he is building an interface between LLMs and your digital life in a macOS environment.

Despite more and more LLM-based apps like editors adopting it, OpenAI remains hesitant, most likely because it's a proposal from a competitor. But as we saw two issues ago, if an ecosystem forms around a technology, like it's currently happening with MCP, not just due to the clients that support it, but the number of integrations that are emerging and the marketplaces that are available, neglecting it or attempting to fork it is most likely a bad idea. For instance, I wonder what Apple will do with Siri. They are now on the spot for not being able to meet their promises around Apple Intelligence. Siri, like Claude and OpenAI models, interacts with users through text or voice-converted-to-text. However, while Claude suggests a standard method to maintain conversation context (MCP), Siri uses its own proprietary interface instead. Will they adopt MCP between Siri and the system? Will they disregard MCP (and its ecosystem) and continue betting on a proprietary solution? We'll see in this year's WWDC, but it's an excellent opportunity for Apple to jump on the existing ecosystems and find ways where Apple's hardware and OS can augment the possibilities, as opposed to neglecting it.

On the app development side, this whole trend also poses a very interesting question. If the interface becomes more text-based, and Siri (or its future evolution) takes that role across all Apple devices, and apps become a context provider, the UI of the app will become even less important. So all the heavy investment teams are doing these days in coming up with a perfect UI and an architecture that powers it might not be as relevant anymore. But even more than that, if all the application does is providing context, and many apps are presentation layers to server services, what's the point in adding the indirection of an app, if Siri could proxy directly with the server that exposes the interface. Sure, from the financial perspective it makes sense because you force developers through your marketplace rules, but it'll create unnecessary friction worsening the experience. Imagine having to install an app to just interface with your Linear issues (as opposed of authenticating against your Linear account using web technologies).

Who knows what the future holds, but languages are here to stay and challenge many of the patterns and assumptions that exist since the inception of the Internet and app development. I'm personally excited because it means new opportunities for companies and organizations, and also more accessible technologies. So if you are building an app, I'd recommend to keep an eye on what's happening and looking at the patterns that are emerging with curiosity. Sooner or later, you might need to design a language interface for your app, so that language-based apps can interface with it.

Food for thought