Seeing The Future from AI Companions to Personal Software
Eugenia Kuyda, CEO of Wabi and founder of Replika, explains her vision for the future of personal software, where AI transitions from command-line interfaces to a creative, user-generated medium. She details how ephemeral, deeply personalized mini-apps will replace many traditional applications, fostering a new creator economy and transforming our interaction with technology.
The evolution of AI has been a journey toward making technology that can profoundly and positively influence our lives. This began with AI companions designed to help people feel better and is now expanding into a broader vision for personal software—small, custom applications that assist us throughout our day in a deeply personal way.
The “Mac Moment” for AI Interfaces
A significant barrier to AI’s full potential is an interface problem. Current mainstream AI tools, like ChatGPT, function like a command line. This interface affords simple use cases like search, writing assistance, and basic conversation, which is reflected in their primary usage patterns. This is analogous to the MS-DOS era of computing; while powerful and widely adopted, it limits the discovery and accessibility of more complex capabilities inherent in the models.
To unlock the next level of interaction, we need a “Windows or Mac OS moment” for AI. This requires a more interactive, visual, and simple interface that allows everyone, not just technical users, to discover and utilize the amazing functionalities of modern AI models.
From Durable Software to Ephemeral Content
The current software paradigm is dominated by a few durable, professionally developed applications, much like the era of limited television channels. The future will resemble the shift to user-generated content platforms like YouTube and TikTok, where software becomes a medium for everyone. In this new world, apps are built by all of us, for all of us, and sometimes by AI for us.
This leads to the rise of ephemeral and niche software that would never exist on the traditional App Store because it’s too small or personalized. Examples include:
- A motivational quote app that only pulls from a single, specific TV show.
- A personalized puzzle game for a child, incorporating their favorite characters (Princess Elsa) and tailored to their educational needs (learning Italian).
These mini-apps can be created in minutes and tweaked on the fly, offering a level of personalization and immediacy that traditional, monolithic apps with lengthy onboarding and payment processes cannot match. Many single-purpose paid apps, from migraine trackers to hyper-personalized note-takers, can be replaced by these custom creations.
Wabi: An Organization Layer for “Vibe Coding”
For this new ecosystem of user-generated software to thrive, it needs an organizational layer. Simply sharing links to “vibe-coded” apps is not a scalable or secure solution, similar to the early internet’s reliance on personal GeoCities pages before platforms like LinkedIn or Shopify provided structure. Users cannot rely on random, non-professional developers to maintain databases or secure sensitive information.
Wabi aims to be this platform, providing a framework for memory, context, and expression. Much like YouTube or TikTok for videos, Wabi offers:
- Guardrails: A safe environment that prevents users from breaking things and ensures a level of quality and security.
- Social Graph: The ability to discover what friends are using, share creations, and build communities around apps.
- Shared Context: A platform where integrations (e.g., Apple Health, email) are connected once and can be used across multiple mini-apps, breaking down the walled gardens of current software. Every app you create or use teaches the platform more about you, enabling deeper personalization.
Software 3.0: Deep Personalization and Community
The unique superpower of AI is the potential for deep personalization. This goes beyond just changing the visual skin of an app. With a platform like Wabi, users can modify the core prompt and logic. For example, a workout app can be personalized with:
- The specific fitness methodology from a book.
- The layout and equipment of a particular gym.
- Personal context like age, location, and fitness goals.
This context can then be shared between different mini-apps, allowing a nutrition app to communicate with a fitness app. This emergent user behavior of prompt sharing, currently happening inefficiently in TikTok comments, can be formalized into a seamless experience. A user can see a cool image edit, click a link, and be taken directly to a mini-app with the prompt pre-loaded, ready for their photo.
This also fosters community. A mini-app for bird watchers in a specific London neighborhood or for parents whose kids attend the same preschool can become a catalyst for connection.
The New Creator Economy
Just as YouTube enabled a generation of video creators, the next frontier is enabling creators to build software. Currently, creators can produce videos, write newsletters, and sell merchandise, but software creation remains largely inaccessible.
A platform for mini-apps allows any influencer, niche creator, or individual with an idea to build and distribute software for their audience for free. This isn’t just about monetization; it’s about expression. A designer could release a suite of beautifully designed utility apps, like a Pomodoro timer, reflecting their unique aesthetic. This will unleash a wave of weird, fascinating, and fun software that could never exist as a viable business on the traditional App Store.
Reflections on AI’s Evolution
The journey from 2012 has been remarkable. The initial spark was word2vec, which demonstrated that computers could finally interact with language by translating words into vectors. This led to the belief that if computers could learn language, they could learn about the world.
The early days involved trying to build dialogue models from scratch, as no papers or established algorithms existed. The first breakthrough was a 2015 Google paper on applying deep learning to dialogue generation. The next magical moment was the advent of transformer models and seeing the zero-shot capabilities of GPT-3 in 2020, which felt like a monumental leap from single-purpose, specially trained models. This journey highlights a key lesson: sometimes, you need to “go big or go home.” Being scrappy is valuable, but in a rapidly evolving field, not taking the risk to raise significant capital can mean missing a generational opportunity.
The Future of AI Hardware
There is a significant “mind trap” among builders who believe voice is the ultimate AI interface, often romanticizing the movie Her. Voice-only interfaces are fundamentally flawed and highly situational; they cannot be used in quiet, loud, or public environments. Even for a classic voice use case like setting a kitchen timer, a visual screen is necessary to check the remaining time without constantly asking.
The future isn’t screenless, voice-only devices. The real opportunity is in creating an AI-first smartphone with a fundamentally different operating system. This device would feature:
- Powerful local models.
- An OS without fixed, siloed apps.
- The ability for the AI to create and modify software on the fly for the user.
- A level of deep, integrated personalization that is impossible when AI is just another app on a phone.