a16zSeeing The Future from AI Companions to Personal Software
At a glance
WHAT IT’S REALLY ABOUT
Wabi’s vision: AI mini-apps as personalized, social software content platform
- Kuyda argues today’s chatbots are a “DOS-era” AI interface that funnels people into basic use cases like writing and search, signaling an interface discovery problem rather than a model capability problem.
- Wabi proposes a future where software becomes malleable and personal—users and AI generate niche, short-lived “ephemeral” apps for specific moments (e.g., a trip, bedtime games) alongside durable everyday tools.
- The platform thesis is “software as content”: mini-apps are shareable, remixable artifacts (like YouTube videos) that can spread through social graphs, comments, and creator communities rather than through traditional app stores.
- A core differentiator is an “organization layer” with guardrails—hosting apps, data, integrations, and shared context so non-developers can safely build and use software without insecure link-sharing or managing infrastructure.
- Kuyda connects this to “Software 3.0”: deep personalization emerges from shared memory and context across apps, enabling workflows where mini-apps learn the user and coordinate with each other over time.
IDEAS WORTH REMEMBERING
5 ideasChatbots concentrate usage because their interface hides AI’s breadth.
Kuyda notes most people use ChatGPT-like tools for writing and search, not because models can’t do more, but because a plain text box doesn’t guide discovery—similar to a command line’s limited affordances.
The next AI breakout may look like a “GUI moment,” not a model moment.
She frames today’s chat UIs as the DOS era and predicts a Windows/macOS-style interface layer that makes powerful behaviors obvious through interactive, visual mini-apps.
AI enables “ephemeral software” that’s too niche to justify App Store economics.
Examples include a hyper-specific motivational quote app or a bedtime puzzle game customized to a child’s favorite characters and language—built in minutes and tweaked continuously.
Creation will be power-law distributed, but tweaking will be mainstream.
Kuyda expects <10% to create from scratch, while many more will remix and request changes via comments—mirroring consumer creation dynamics seen in UGC platforms.
A trusted platform layer is required for UGC software to be safe and durable.
She argues link-shared vibe-coded apps are risky (e.g., data leaks from amateur-built apps) and that users need guardrails: hosted databases, permissions, and a stable environment to store personal data.
WORDS WORTH SAVING
5 quotesWe felt like there must be an interface problem. Uh, when people look at a command line or a chatbot, they really just see, um, you know, search, writing tool.
— Eugenia Kuyda
Think of it as an operating system built on the platform of you and not on some random, um, fixed-... context.
— Eugenia Kuyda
Oh, that was the most... And I guess this is why we started the company. Um, g- how is it possible that we have this godlike technology, yet we pass around these text prompts, which is almost like Microsoft DOS commands, but worse.
— Eugenia Kuyda
Sometimes you need to sorta go big or go home. And not having the balls to do that, especially in this current environment, I think you can suffer the consequences.
— Eugenia Kuyda
There's a huge mind trap that exists among, uh, builders in the space where they somehow think that voice is the main interface.
— Eugenia Kuyda
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome