Why connected agents aren’t truly collaborating yet

AI must evolve from mere connectivity to genuine collective intelligence

Right now, most AI systems can connect, but they don’t actually think together. Vijoy Pandey, SVP and GM of Outshift by Cisco, put it clearly, today’s agents talk to each other in basic ways, but there’s no shared reasoning or understanding. Connectivity protocols like MCP, A2A, and AGNTCY were critical first steps. They allow data exchange and communication, yet they operate on a superficial level. What we need now are systems that share intent, context, and purpose. That’s where the next generation of AI needs to go.

Business leaders should view this shift as an evolution from information transfer to collective intelligence. Real progress will come when AI agents can align goals, interpret context, and collaborate toward outcomes, much the way high-performing teams work together. These systems won’t just share information; they’ll process it together, evaluate it, and adapt their actions cooperatively.

This approach has strategic implications. In complex organizations, especially those operating globally, moving past connectivity into shared cognition will create intelligent workflows that are dynamic, context-aware, and continuously learning. It will reduce inefficiencies caused by fragmented communication between systems and make human-AI collaboration far more seamless.

Vijoy Pandey captures this simply: agents today “can connect together, but they can’t really think together.” That gap defines the next big opportunity in enterprise AI systems. The companies that close it first will set the pace for the next wave of innovation in distributed intelligence.

Shared intent and knowledge form the bedrock of collective innovation among AI agents

To move AI forward, we need to take a lesson from human evolution. Human intelligence became collaborative when language allowed shared goals and knowledge. Around 70,000 years ago, language allowed people to exchange information and to align intentions and build on each other’s ideas. That’s what led to collective innovation, and it’s the capability AI systems are still missing.

Both Vijoy Pandey and Noah Goodman, professor at Stanford University and co-founder of Humans&, pointed out that collaboration among agents requires the same conditions: understanding intent, sharing knowledge, and evolving context. Goodman notes that language is more than encoding and decoding, it’s about understanding the speaker’s purpose and the world around them. In AI, that means enabling systems to grasp not just data but the motivation behind it, the situational context, and how it influences the next action.

For executives, the message is straightforward. The companies investing in agents that understand purpose and context, not just commands, will unlock breakthroughs in innovation. When AI systems can share intent and adapt knowledge in real time, they move from reactive assistants to proactive problem solvers. That’s when collaboration scales beyond individuals and departments, across teams, ecosystems, and even industries.

Vijoy Pandey warns that current models are stuck at coordination, while Goodman explains that genuine collaboration requires context-aware cognition. Together, their insights highlight where the real edge lies, developing AI that doesn’t just connect data but connects understanding.

The proposed “internet of cognition” architecture aims to enable multi-agent collaboration

Vijoy Pandey, SVP and GM of Outshift by Cisco, proposed what he calls the “Internet of Cognition” — a new framework to help AI agents move from isolated intelligence into truly collective operation. It’s structured around three layers that together form the foundation for what he describes as shared cognition across diverse systems.

The first layer, the Protocol Layer, standardizes how agents communicate, share intent, and negotiate outcomes. It’s more than message passing; it’s about establishing a common ground for purpose and understanding across networks and vendors. The second, the Fabric Layer, acts as shared memory. It allows agents to retain, modify, and evolve context based on new data and experience. The last, the Cognition Engine Layer, governs the pace and control of intelligent behavior, it enables faster reasoning but ensures it remains compliant with policy, secure against attack, and within cost parameters.

For executive leaders, this structure signals a shift away from isolated deployment toward interconnected systems that can grow their understanding collectively. Companies adopting this approach will see greater adaptability in decision support, smoother knowledge transfer between departments, and stronger compliance automation. The biggest advantage lies in scalability, each layer contributes to deeper collaboration without sacrificing control or visibility.

Pandey underscores this direction clearly: “We have to mimic human evolution. In addition to agents getting smarter and smarter, we need to build infrastructure that enables collective innovation.” The Internet of Cognition is that infrastructure, one designed to make AI cooperation reliable, auditable, and useful across all business domains.

Redesigning foundation model training is key to enhancing human–AI and multi-agent collaboration

Noah Goodman, Stanford professor and co-founder of Humans&, explained that most foundation models today are built to handle short, linear tasks. They perform well within boundaries but struggle with continuity and extended context when collaboration deepens. At Humans&, his team is taking a different approach, training models through continuous, long-horizon interactions that emphasize relationships, understanding intent, and achieving long-term results across multiple human and agent participants.

Goodman’s goal is not to create more autonomous agents but more collaborative ones. These agents understand roles, expertise, and responsibilities. They know “who knows what” and can connect the right specialists at precisely the right time. This design transforms the way humans and AI systems collaborate, turning AI from a reactive processor into an active participant in complex projects.

For executives, this focus on long-term interaction and collaboration changes how AI supports teams. It enables systems that can track evolving project contexts, retain organizational history, and coordinate independently across business units, all while respecting human direction. In practice, it means improved strategy execution, faster iteration cycles, and more resilient decision-making structures.

Goodman summed it up directly: “Our goal is not longer and longer autonomy. It’s better and better collaboration.” His team’s work at Humans& reinforces a critical truth, AI that understands humans deeply will outperform AI built for autonomy alone. That shift in design philosophy marks a decisive turn toward the future of practical, human-aligned intelligence in enterprise environments.

Establishing adaptive guardrails is essential to balance innovation with control in AI systems

Managing intelligent systems at scale requires guardrails that maintain safety without limiting creativity. As AI becomes integrated across every business layer, from operations and logistics to customer engagement, it will handle tasks previously reserved for expert teams. Both Noah Goodman, Stanford professor and co-founder of Humans&, and Vijoy Pandey, SVP and GM of Outshift by Cisco, addressed this growing challenge: designing boundaries that protect trust while still supporting intelligent adaptability.

Goodman pointed out that humans don’t operate purely on strict, rule-based systems. Instead, they consider context, intent, and likely consequences before acting. Translating that principle into artificial systems means developing adaptive guardrails that allow AI to make nuanced decisions without violating standards. Pandey added that implementation must be collaborative, where teams contribute to defining flexible limits through iteration and shared understanding.

For business executives, this is a governance question as much as a technical one. A rigid control framework stifles initiative, while insufficient oversight leads to risk. Adaptive guardrails balance both, maintaining compliance, ensuring operational transparency, and enabling innovation to continue safely. They also allow rapid adjustment, ensuring AI can evolve alongside shifting regulations, technologies, and organizational objectives.

Goodman framed the central problem precisely: “How do we provide the guardrails in a way which is rule-like, but also supports the outcome-based cognition when the models get smart enough for that?” The path forward is clear, companies need intelligent governance systems that allow contextual decision-making while maintaining measurable accountability.

Distributed intelligence, integrating human and AI efforts, is the future pathway to superintelligence

The next leap in artificial intelligence won’t come from larger individual models but from highly connected, distributed networks that merge human insight with AI cognition. Vijoy Pandey emphasized this during the AI Impact Series, explaining that progress will follow two trajectories, vertical development, which enhances the capabilities of each agent, and horizontal expansion, which strengthens the cooperative network between agents. When both progress together, intelligence compounds across the entire system.

Noah Goodman reinforced that these systems must remain human-integrated. He warned against creating autonomous AI ecosystems that operate in isolation. The ultimate goal is a distributed intelligence framework where human and machine capabilities align, reinforcing each other. This structure helps humans stay meaningfully engaged, while AI contributes analytical power, memory, and precision at scale.

For executive leaders, this approach encourages strategic investment in connected architectures rather than siloed systems. The reward is exponential leverage, each node in the network strengthens collective performance. Businesses operating within such ecosystems will see higher adaptability, faster decision-making, and more consistent alignment across departments.

Pandey concluded that “true super intelligence will happen through distributed systems,” reflecting a clear direction for enterprise AI strategy. Goodman’s perspective complements it: A future where “there’s an integrated ecosystem, a distributed ecosystem that seamlessly merges humans and AI together” represents sustainable innovation. For global organizations, it means pursuing an intelligence model that is connected, transparent, and firmly anchored in human purpose.

Key highlights

• Evolve from connection to cognition: Connectivity alone no longer drives value. Leaders should invest in systems that enable AI agents to reason collectively, aligning goals and context rather than just exchanging data.
• Enable shared intent and knowledge: True innovation emerges when agents, like teams, share purpose and contextual understanding. Executives should focus on AI infrastructures that synchronize intent across departments and ecosystems.
• Adopt layered architectures for collaboration: The Internet of Cognition model offers a clear blueprint. Organizations should implement layered systems, protocol, fabric, and cognition layers, to enable scalable, secure, and context-informed collaboration.
• Redefine training for long-term human–AI synergy: Foundation models must be trained for deep, continuous collaboration, not just independent performance. Leaders should support training approaches that build socially and contextually aware AI.
• Implement adaptive guardrails to balance control and creativity: Governance must shift from rigid rule enforcement to flexible frameworks. Executives should design oversight systems that maintain compliance while allowing intelligent, context-driven decisions.
• Invest in distributed intelligence ecosystems: The future of AI lies in networked systems that merge human expertise with collective machine intelligence. Decision-makers should prioritize connected, human-aligned architectures that scale intelligence horizontally and vertically.