Getty
The enterprises that built on open, interoperable standards in the 1990s became the infrastructure of the digital age. The ones that built walled gardens were eventually dismantled, acquired or made irrelevant.
This pattern could be about to repeat itself, and the window to be on the right side is narrowing fast.
"The takeaway from the 1990s is not simply 'open won,'" says Vijoy Pandey, general manager and senior vice president of Outshift by Cisco. "The takeaway is that infrastructure decisions made during a narrow window of time became permanent. … Once the wiring was in, you couldn't rip it out and start over."
In AI, the protocols, identity systems, discovery services and shared memory layers that enterprises adopt in the next 12 to 18 months will become foundational for multi-agent software. With the right infrastructure in place, enterprises can build toward what Pandey and his team at Outshift call the Internet of Cognition, which enables agents to share intent, accumulate knowledge and innovate collectively in an open and interoperable way— rather than just connect and communicate.
"The 1990s internet solved a connectivity problem," Pandey says. "What we are building now is harder because we are solving a cognition problem. Agents don't just need to exchange opaque data, they need to peek inside and build shared understanding, align on goals and accumulate knowledge over time."
Most enterprises today deploy agents the way they did with early web applications: as standalone services that handle specific tasks for specific departments. A site reliability engineering agent here, a coding assistant there, a customer service agent in support, a finance agent over in financial planning and analysis. Each one works fine in its own lane.
"When you need those agents to coordinate on a complex business outcome … they can’t," Pandey says. "They don’t share intent, they don't share context and they definitely don't build on each other's knowledge. Just like human teams, they need to coordinate and negotiate, and compromise and adjust."
Those enterprises that will define the AI economy are making different architectural choices. They're building systems designed from the start to work across departments, vendors and organizational boundaries.
The goal isn't a collection of capable agents. It's a network of agents that gets smarter as it grows.
"Build for composition, not isolation," Pandey says. "That's the single most important architectural principle right now."
The travel industry figured this principle out decades ago, with several industry leaders opting to build interoperable systems that allowed airlines, hotels and rental car companies to compose services together. That early architectural decision is why booking a complex multi-leg international trip with a hotel and a rental car now takes 30 seconds.
"The industries that invest in composable infrastructure early always capture the value," Pandey says. "The ones that built proprietary silos spend decades trying to unbundle themselves or pay a steep price."
Some enterprises may worry that committing to open, interoperable standards means ceding competitive advantage to rivals building on the same foundation. Pandey frames that as a false trade-off.
"Open infrastructure actually protects your proprietary investment because it means your agents aren't locked into a single vendor's ecosystem," he says.
In the age of multi-agent AI, vendor lock-in can become an existential threat. Enterprises that build on proprietary agent infrastructure are one pricing change or API deprecation away from seeing their systems siloed and stranded.
The Internet of Cognition is designed to amplify, rather than replace, human judgment.
"We are building an architecture for multi-agent-human teams," Pandey says. "Humans are first-class participants, not supervisors watching dashboards."
A multi-agent system without human participation optimizes within known boundaries, whereas a human-agent network can reason through problems that neither could solve alone. This is precisely the kind of capability that will distinguish obsolete infrastructure from decade-defining innovation.
"The path to distributed artificial superintelligence runs through collaboration between agents and humans," Pandey says. "Agents operate at machine speed and scale. Humans bring judgment about genuinely novel situations [and] ethical reasoning."
In moments of technological uncertainty, it's natural to feel overwhelmed by the options. Especially with AI, analysis paralysis is real. Here, Pandey has seemingly counterintuitive advice: Don't agonize over which particular AI model you adopt.
"Models are getting better and cheaper on a curve that won't slow down," he says. "Your differentiation will never be which model you picked."
Instead, focus on what you're building around them. With agentic AI, durable advantages will come from open infrastructure that isn't hostage to a single vendor's roadmap; institutional knowledge that deepens over time; and governance built into the foundation rather than retrofitted afterward.
For business leaders making the difficult decisions about their company's AI future, Pandey has three recommendations:
What it comes down to, Pandey says, is whether your organization intends to define the architecture or rent it from someone else.
"The question for every enterprise is whether they want to help shape how collective intelligence works or adopt whatever someone else decides to build. If the history of computing teaches us anything, the shapers capture disproportionate value."
To learn more about the Internet of Cognition, read "From Connection To Cognition: Scaling Out Superintelligence," or explore the concept interactively at Outshift by Cisco.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。