humans&
humans& is a human-centric frontier AI lab building models that understand people and help them work together. Its stated aim is to make AI strengthen trust, connection, organizations, and communities rather than simply automate work away.
At a Glance
- Consumers and families
- Friends and community group chats
- Developers building collaborative AI agents
- Education and tutoring
- +2 more
AI Tools by humans&
(1)Boba
Text Based AI Personal Assistant
Discussions
No discussions yet
Be the first to start a discussion about humans&
Latest News
Published GPU-accelerated Weight Teleportation, a lossless GPU-compression approach for faster reinforcement-learning weight synchronization.
Introduced &Tutor (Ampersand Tutor), a tutoring model trained with Persimmon and made available through the API.
Published Faster-than-light NVFP4 decode—with precision, describing a higher-precision NVFP4 decode approach for inference and RL rollouts.
Unveiled Boba, a social intelligence for personal and group chats that understands goals and communities.
Products & Services
A free, chat-based social intelligence assistant that works one-to-one or in group chats. It handles reminders, calendars, email triage and drafting, web research, travel and restaurant planning, bookings, shared family logistics, images, and songs; it is available via iMessage, SMS/RCS, and web.
A collaborator model trained for people to achieve individual and collective goals. It supports long-horizon collaboration, social context, proactive assistance, conflict handling, multi-agent coding, and explicit send_message/pass tools through the Agents API.
A large-scale trained user model and research preview used to simulate people’s goals, preferences, conversation behavior, and changing feedback for training and evaluating collaborative models.
A model trained to teach by training against Persimmon as a simulated learner; it is offered through the Ampersand API and targets adaptive tutoring and durable student understanding.
Market Position
humans& positions itself against the dominant task-completion and autonomous-agent paradigm by optimizing for human-AI complementarity, social understanding, long-term goals, and group coordination. Ampersand is differentiated from general-purpose frontier models such as OpenAI, Anthropic, Google, and xAI offerings through its simulated multi-user training and collaborator-specific evaluations; Boba competes with general AI assistants and task agents by operating inside existing conversations and speaking only when useful.
Leadership
Founders
Eric Zelikman
CEO and co-founder; previously an early researcher at xAI and a Stanford computer-science PhD candidate advised by Nick Haber and Noah Goodman. He co-authored STaR and Quiet-STaR.
Andi Peng
Co-founder; previously a research scientist at Anthropic working on reinforcement learning and post-training for Claude, and a Special Government Employee at the U.S. Defense Innovation Unit.
Georges Harik
Co-founder; Google’s seventh employee, who helped build Google’s early advertising systems and contributed to products including Gmail and Google Docs.
Yuchen He
Co-founder; previously an xAI researcher involved in developing the Grok chatbot.
Noah Goodman
Co-founder; Stanford professor working across psychology and computer science, and a researcher whose prior affiliations include Google DeepMind.
Executive Team
Eric Zelikman
Co-founder and Chief Executive Officer
Former xAI researcher and Stanford computer-science PhD candidate; co-author of STaR and Quiet-STaR.
Founding Story
The company was founded in September 2025 by researchers and builders from Anthropic, xAI, Google, OpenAI, Meta, Stanford, and other AI labs. The founders’ initial vision was to build AI around how people connect and work together, using long-horizon and multi-agent reinforcement learning, memory, and user understanding so that humans and AI remain complementary.
Business Model
Revenue Model
API usage charges for Ampersand models based on input, cached-input, and output tokens; Boba is offered as a free consumer service with no subscription and no ads.
Pricing Tiers
The public ampersand-v0.1 and ampersand-tutor-v0.1 models use these rates.
30 API requests/minute, 10 work requests/minute, 16 concurrent executions, and 10,000 sessions.
600 API requests/minute, 300 work requests/minute, 64 concurrent executions, and 1,000,000 sessions.
Free to use; no subscription, no ads, and no download required.
Target Markets
- Consumers and families
- Friends and community group chats
- Developers building collaborative AI agents
- Education and tutoring
- AI research and model-evaluation teams
- Organizations seeking human-centered coordination and productivity tools
- Helping families and friends coordinate plans, chores, appointments, trips, and recurring activities
- Personal accountability, reminders, health routines, and long-running goals
- Email triage, drafting, web research, travel planning, restaurant discovery, and bookings
- Group-chat facilitation and resolving disagreements
- Developer-built collaborative agents and multi-agent workflows
- Adaptive tutoring and education
- Boba early users, who had exchanged more than 3 million messages by the October 2026 product page; more than 300,000 were in group chats.