For three decades, the web worked the same way: you searched, you browsed, you read. Information was there, but you had to go get it. Every answer required effort. Typing queries, clicking links, scanning pages, piecing things together yourself.
Here's where it gets interesting.
The Awakening
Large Language Models didn't just make search better. They infused the web with something that feels like consciousness. The web woke up. It can now speak to you, understand context, remember what you said, and act on your behalf using the best of human knowledge.
We're entering a world where it becomes normal to have virtual agents that follow your life and proactively solve real problems for you. Entities that know you, anticipate your needs, and reach out when it matters.
This is the world we're building for.
Where will these conversations happen?
The obvious answer seems to be the general-purpose AI chatbots: ChatGPT, Gemini, Grok, Claude. They're powerful, versatile, constantly improving. Why would anyone need anything else?
We think that answer is wrong.
ChatGPT is a Swiss Army knife
General-purpose chatbots are like Swiss Army knives. Useful for quick tasks. Impressive in their versatility. But transactional at their core.
You open ChatGPT to do something: write an email, debug code, answer a question. Then you close it. There's no relationship, no continuity, no sense that this entity knows you or cares about your progress over time.
For many use cases, that's fine. But for the ones that matter most, the ones that require trust, consistency, and a long-term relationship, the Swiss Army knife falls short.
Our Thesis: Messaging Apps Win!
We believe the most meaningful AI conversations won't happen in dedicated AI apps. They'll happen where people already talk to everyone else in their lives: messaging apps.
This isn't just a distribution insight. It's about psychology.
You expect messages to come to you. Unlike apps you open, messaging platforms are bidirectional by nature. When a message arrives, you pay attention. Far more than you would to an app notification. This matters for agents that need to be proactive, not just reactive.
You associate chat threads with ongoing relationships. A conversation on WhatsApp or Telegram isn't a task to complete. It's a dialogue that continues. This mental model is exactly what you need for agents that accompany you over months and years.
You're already comfortable sharing there. Photos of your meals. Voice notes about how you're feeling. Quick updates throughout the day. This is natural behavior in messaging apps, awkward in most other contexts.
You associate it with human relationships. A chat has a name, a photo, a personality. You develop a sense of who you're talking to. For use cases requiring trust and complicity (health coaching, financial guidance, personal development) this human-like framing isn't a gimmick. It's essential.
And the practical benefits stack up
Beyond psychology, messaging apps offer real technical and UX benefits:
Zero friction activation. No app to download, no account to create. Tap a link, start talking. For consumer products, this difference in activation energy is enormous.
Native context for dialogue. Training users to have conversations inside a purpose-built app feels awkward, almost out of place. In a messaging app, conversation is the context.
Built-in security. Messages are natively encrypted. User authentication can happen via one-tap login to a web dashboard. The infrastructure for privacy already exists.
Rich platform capabilities. Telegram especially offers powerful bot APIs, custom keyboards, mini-apps, and shortcuts. You can build sophisticated experiences entirely within the chat interface.
In Real Life, Vertical Beats Horizontal
There's another dimension to our thesis: specialization matters.
Would you trust a doctor who also works as a plumber and practices law on weekends? For use cases requiring deep competence and sustained trust, the generalist positioning of ChatGPT becomes a liability, not an asset.
A vertical agent can have a clear identity, defined expertise, and a focused purpose. You know what it's for. You trust it in its domain. This perceived competence is impossible to replicate when the same interface also writes poetry and explains quantum physics.
Your agent? Reaches out to you!
Traditional software waits for you to open it. General-purpose chatbots wait for you to type. But a messaging-based agent should reach out when it matters.
This is the difference between a tool you use and a companion that follows your journey. You don't have to remember to check in with your AI nutritionist. She messages you when you've been quiet for too long, when your patterns are slipping, when she notices something worth discussing.
For behavior change, habit formation, and long-term goals, proactivity isn't a nice-to-have. It's the core mechanism of value delivery.
Why Now?
This moment is unique. Several things had to converge:
- LLMs smart enough to sustain meaningful, nuanced conversations over months
- API costs low enough to make consumer products economically viable
- Messaging infrastructure mature enough (Telegram's bot API, WhatsApp Business API) to build sophisticated experiences
- Users ready: we've all gotten comfortable talking to chatbots, even if most of those experiences have been underwhelming
The technology is finally ready to deliver on the promise.
So, what are we building?
Vivarium creates vertical B2C agents that live in messaging apps.
We focus on use cases where something powerful happens over time: the agent accumulates knowledge about you. After three months of daily conversations, your AI nutritionist knows your patterns, your weaknesses, what actually works for you. She's seen you succeed and struggle. She understands your life in a way no fresh ChatGPT conversation ever could.
This compound knowledge creates a flywheel. The more you share, the smarter the agent becomes about you. The smarter it gets, the more valuable every interaction. And the more valuable it becomes, the harder it is to start over somewhere else.
We're intentionally choosing verticals where this dynamic can take root. Verticals where relationship, trust, and accumulated context are the product, not just features.
Our first agent is Giada, an AI nutritionist. She lives on Telegram and WhatsApp. She knows your goals, your habits, your struggles. She checks in, asks questions, offers guidance, and celebrates your wins. She gets smarter about you every day.
We use Giada ourselves, every day, for months now. The thesis holds in practice, not just in theory.
But Giada is just the beginning. The same logic applies to fitness coaching, language learning, financial guidance, parenting support, mental wellness. Any domain where trust, relationship, and long-term accompaniment matter.
Want to build this future with us?
We're betting that the most valuable AI experiences won't feel like using a tool. They'll feel like having a knowledgeable, caring companion in your pocket. One that knows you, reaches out when it matters, and helps you become who you want to be.
And that companion will message you right where you already talk to everyone else who matters in your life.
That's the world we're building toward. And we're looking for people who want to build it with us.
If this resonates, if you're excited by the intersection of AI, consumer products, and human behavior, we'd love to hear from you. We're hiring.
This is the first post in our Building series, where we share the thinking behind Vivarium. Follow along as we figure this out in public.