
Who says AI chatbots can't be social?
As conversational AI becomes more advanced, developers are increasingly focusing on making interactions feel natural, engaging, and emotionally aware. Modern AI systems can adapt their tone, remember context throughout a conversation, and respond in ways that feel less like a search engine and more like a thoughtful discussion partner.
The next frontier isn't just intelligence. Instead, it's kind of like social intelligence, wherre virtual assistants can be designed to communnicate more like people do, and socialize with nuance, personality, and an understanding of conversational dynamics.
Pika Labs is aiming for that, with the introduction of a new feature called 'Group Chats with Agents.'
Available directly within its platform at pika.me and the Pika Agent iOS app, the update allows users who have created a Pika Agent to bring that agent into shared conversations with friends, family, or colleagues.
Rather than limiting interactions to one-on-one chats, the system now supports group settings where the AI can participate alongside human users.
To use the feature, a user starts a new group chat in the web interface or iOS app, names it, and shares a link to invite participants.
Once everyone is in the chat, mentioning the agent by its handle prompts it to join the conversation.
The agent then responds to messages, suggests ideas, and performs tasks based on the group's discussion. Multiple agents can join the same chat, though only the owner of each agent can invite them. The owner also covers any token costs associated with the agent's actions or generated content.
The demonstration video accompanying the announcement shows several everyday scenarios.
In one family chat titled "The Jones Zone," a user introduces the agent Zoltar and asks for restaurant recommendations after noting a preference for gluten-free options. The agent quickly replies with two nearby suggestions, complete with images and details like distance and menu highlights.
In another example, a team working on a game show project requests that the agent Melanie add confetti effects to a prize reveal animation.
The agent generates an updated image featuring a chainsaw on a pedestal against a colorful rainbow backdrop with falling confetti, and the group reacts positively in the chat.
A third quick clip depicts a request for a birthday video message, illustrating how agents can contribute media directly into the conversation flow.
The chat is about to pop off. Today, we’re launching the first in-app Group Chat with an agent.
Have your Pika Agent help your parents figure out their new iPhone update. Make memes with your friend group. Or collab on a microdrama with your teammates (honestly, why not?) pic.twitter.com/FuO0yBruie— Pika (@pika_labs) June 4, 2026
According to details on the pika.me site, agents in these group chats can handle a range of practical and creative tasks.
For families, this might include booking reservations, scheduling vacations, setting event reminders, or making appointments. Friend groups could use an agent for generating birthday videos, creating photo collages, or planning outings. In professional contexts, agents might assist with content creation, coding support, or other collaborative projects.
The agents draw on their customizable personalities, persistent memory, and multi-modal capabilities, supporting text, images, video, and more, to contribute without disrupting the natural back-and-forth of the chat.
The feature remains experimental, as noted on the platform. Agents may not always interpret intentions perfectly on the first attempt, and users are encouraged to refine prompts through trial and interaction.
Pika Agents are not designed for providing medical, legal, or financial advice, and the platform includes moderation tools along with age restrictions for users.
This addition builds on Pika Labs' existing focus on customizable AI agents that go beyond simple chatbots, integrating them more directly into collaborative spaces. By enabling agents to join group discussions, the platform positions them as active participants rather than standalone tools.