AI's Next Frontier: Building Models for Human Coordination

The artificial intelligence landscape is witnessing a strategic pivot. While recent years have been dominated by models excelling at conversational tasks, a new cohort of researchers is looking toward a more complex horizon: collaborative intelligence.

A new startup, Humans&, is at the forefront of this movement. Founded by alumni from some of the most prominent names in AI research, the company is building the next generation of foundation models with a singular focus on coordination rather than chat.

This shift addresses a fundamental limitation in current AI systems. While they can process information and generate text, orchestrating multiple agents to work together on complex, real-world problems remains largely unsolved. The venture aims to bridge this gap.

The founding team behind Humans& brings together an unusual concentration of expertise. The startup was established by alumni from five of the world's leading AI research institutions: Anthropic, Meta, OpenAI, xAI, and Google DeepMind.

This collective background represents a cross-section of the industry's most advanced research into large language models and safety protocols. The team's experience spans the development of foundational models, alignment research, and the infrastructure required to train massive AI systems.

By drawing talent from these competing labs, Humans& has assembled a team with deep insights into both the technical architecture of modern AI and the practical challenges of deploying these systems at scale.

• Expertise in large-scale model training
• Deep knowledge of AI safety and alignment
• Experience with multi-agent systems
• Understanding of real-world deployment challenges

The core thesis of Humans& is that the next major leap in AI capability will not come from better single-turn conversations, but from systems that can effectively coordinate with each other and with humans.

Current foundation models are largely designed as autonomous agents that respond to prompts. The new paradigm being explored focuses on models that can understand group dynamics, manage shared goals, and facilitate complex workflows between multiple participants.

This approach requires a fundamental rethinking of how AI models are trained and evaluated. Instead of optimizing for individual response quality, the focus shifts to measuring collective outcomes and the efficiency of collaborative processes.

Building models for collaboration, not chat.

The implications of this shift are significant. It moves the application of AI from a tool for individual productivity to a potential infrastructure for organizational coordination, project management, and complex problem-solving teams.

Developing foundation models for coordination presents unique technical challenges. These systems must be capable of contextual awareness across multiple agents, understanding not just individual inputs but the relationships between different participants' goals and constraints.

Training such models requires novel datasets and evaluation metrics. Traditional benchmarks for language models focus on accuracy and coherence in isolation, whereas collaborative systems need to be measured on their ability to achieve shared objectives efficiently.

The architecture of these models likely involves more sophisticated memory systems and state management than current chat-focused models. They must maintain awareness of ongoing tasks, individual roles, and collective progress over extended interactions.

• Multi-agent communication protocols
• Shared goal alignment mechanisms
• Dynamic role assignment capabilities
• Conflict resolution in collaborative settings

The emergence of Humans& reflects a broader trend in the AI industry toward specialization. As the foundational capabilities of large language models mature, researchers are increasingly focusing on domain-specific applications and novel interaction paradigms.

This development also highlights the fluid movement of talent between major AI labs. The concentration of expertise from Anthropic, Meta, OpenAI, xAI, and Google DeepMind within a single startup suggests that the boundaries between these organizations are permeable, with knowledge and innovation spreading across the ecosystem.

For the broader technology sector, successful coordination models could unlock new categories of software and services. From automated project management to complex scientific research collaboration, the potential applications span virtually every industry that relies on teamwork and collective decision-making.

Humans& represents a calculated bet on where artificial intelligence will evolve next. By focusing on coordination rather than conversation, the startup is addressing a fundamental gap in current AI capabilities.

The success of this venture will depend on whether the technical challenges of multi-agent collaboration can be solved at scale. If successful, it could redefine how we think about AI's role in organizations and complex systems.

As these models develop, they may shift the industry's focus from individual AI assistants to integrated collaborative ecosystems, marking the beginning of a new chapter in artificial intelligence.