Equity

Eugenia Kuyda on Why AI Agents Should Be Multiplayer

Eugenia Kuyda· Founder & CEO at Wabi
·~29 min·English·TechCrunch
AgentsAI CompanyBusiness Strategy
TL;DR

The founder of Replika says the whole personal-agent race is building the wrong shape, single-player assistants that save you time, when what people actually want is a multiplayer tool that helps them spend time well with the people in their life.

01The Core Bet

Agents Should Bring People Together

Kuyda's whole thesis is that an AI agent should connect you to the people in your life, not become one more reason to sit alone with a chatbot.

we believe that AI agents should bring people together, not give people more reasons to spend time alone with AI.

— Eugenia Kuyda, Equity
Key Insight
Making the agent multiplayer is also a moat: a single-player assistant is easy for any big lab to copy, but a shared graph of your family and friends, and the context living in it, is far harder to rebuild.
02Reframing the Value

Saving Time Isn't the Point

After building Replika, Kuyda concluded that saving time is a weak promise; people would rather spend their time better, on the things and people they care about.

most people actually don't care that much about saving time

— Eugenia Kuyda, Equity
Key Insight
This quietly rejects the premise of almost every agent demo. If the goal is to spend time well rather than do more faster, then the product you build, and the metric you optimize, look nothing like a productivity tool.
03Full Circle

The Idea Is 12 Years Old

Kuyda pitched this same messenger-with-agents concept at TechCrunch Disrupt in 2014, years before large language models existed to make it work.

We built it back then without language models which are really interesting because they didn't really exist.

— Eugenia Kuyda, Equity
Key Insight
The vision did not change; the technology under it did. A founder who spent a decade on this problem before language models existed has a sharper map of what a personal assistant should feel like than teams that started the day ChatGPT shipped.
04Design Choice

A Thread for Each Part of Your Life

Instead of one endless chat, Wabi opens a separate thread for each area of your life, which Kuyda argues keeps the experience organized and manageable as it grows.

I think the downsides are pretty significant actually like if you think about a single threaded experience it gets worse over time.

— Eugenia Kuyda, Equity
Key Insight
A single thread also caps how proactive the agent can be: with everything in one feed, Kuyda says you have to be very selective about what you send there. Separate threads lift that limit, because each nudge, a show, a restaurant, a friend's birthday, lands in the part of life it concerns.
05Where Interfaces Go

The DOS Era of AI Interfaces

Kuyda calls today's single-chat assistants the DOS era of AI, functional but primitive, and bets the winning interface will feel more like a graphical OS, with apps the agent builds for you on the fly.

there will be a Mac OS, Windows, so to say for everyone to use and it's going to feel completely archaic that everything just happen in these single thread chat chats.

— Eugenia Kuyda, Equity
Key Insight
The claim is that chat is a transport layer, not the product. Plenty of everyday tasks, tracking weight or watching a budget, are miserable as pure text and obvious as a tiny app, so the agent that can generate those apps on demand beats the one that can only talk.
06Trust Mechanic

Permissions Live in the Chat

Rather than one global give-the-app-everything switch, Wabi scopes access per conversation: your partner thread can see your calendar and email, a casual thread sees far less.

I think that's a really cool way to solve for shared context without necessarily breaking user trust.

— Eugenia Kuyda, Equity
Key Insight
Putting permissions inside a shared chat also makes them legible: because you are in the thread, you watch exactly what someone asks about you and what the agent answers. Visibility, not just a toggle, is what keeps shared access from feeling like surveillance.
07The Cost Angle

Not Every Message Burns a GPU

Kuyda argues Wabi's shape is cheaper to run: people talking to each other in group chats, and using simple rule-based mini-apps, mostly don't touch an expensive model at all.

you're always burning computer. You're always asking them to do something for you.

— Eugenia Kuyda, Equity
Key Insight
It is a pointed contrast with single-agent apps, where every message is an inference call. If much of the valuable activity is human-to-human or handled by deterministic apps, Wabi's compute bill grows far more slowly than its usage, a real edge in a market where serving cost decides who survives.
08Data & Trust

The Trust Bet

Wabi only works if you connect your email, calendar, even finances, so Kuyda leans on Replika's record: 40 million users, no data incident, and a standing promise never to sell their data.

We made a firm commitment to our users from the very beginning with replica that we will never sell their data.

— Eugenia Kuyda, Equity
Key Insight
Proactivity and privacy pull against each other: the more context the agent holds, the more useful and the more dangerous it becomes. Kuyda's answer is reputational rather than technical, a clean record and a public promise, which raises the stakes of a single mistake, because the whole pitch rests on that promise never breaking.