Why AI's next big leap is collective intelligence

AI agents are proliferating -- semi-autonomous LLM agents that can roam the web and local machines and take actions on behalf of the human who creates and/or uses them. And recently, an experiment with "a social network for ai agents" made a splash (and it turned out that many of the most alarming actions they took -- such as creating their own language -- were the product of human pranksters). But the possibility of concatenating and collaborating agentsis real. Like most of my AI posts, I don't know enough to venture a hard opinion whether this is a good thingh or a bad thing -- but my spidey sense tells me that AI agents acting as collective intelligence could be a powerful leap in a direction nobody yet understands.

The fundamental problem holding back AI progress is that agents can connect, but they can't think together. Vijoy Pandey, GM and SVP of Outshift by Cisco, explains why the missing piece of AI is horizontal scaling.

1. First things first: Why does AI feel stuck right now?

Pandey: Scaling up models will move us toward superintelligence, but progress is slowing. Key technical experts and researchers alluding to this trend are calling out for an "age of research," and the need for the next algorithmic breakthroughs.

Here's what could accelerate the timeline: scaling out, not just up. The collective is exponentially more powerful than the individual. We've been building smarter individual agents while ignoring how they could think together.

• The limitation isn't computing, data or parameters; it's that insights stay trapped inside isolated systems and do not compound.

2. Okay, but: Why does this shift from isolated to collective intelligence matter to leaders today?

Pandey: Because the world's most complex problems — from drug discovery to global supply chain optimization — cannot be solved by a single model or agent. It needs teams of agents, from different vendors and with different expertise, working towards common goals.

• Right now, when one agent figures something out, that knowledge only lives within that agent. Every other agent starts from scratch.

Without infrastructure for collective intelligence, you're deploying isolated experts that can't build on each other's work and cannot leverage each other's expertise.