Building the Internet of Agents with Vijoy Pandey - #737
🎯 Summary
Podcast Summary: Building the Internet of Agents with Vijoy Pandey - #737
This episode features Vijoy Pandey, SVP and GM of Outshift by Cisco, discussing the concept of the Internet of Agents as the necessary platform for future agent-to-agent collaboration, moving beyond current siloed LLM applications.
1. Focus Area
The primary focus is the Internet of Agents (IoA): defining it as an open, interoperable, and quantum-safe platform for enabling complex, multi-agent workflows within the enterprise. The discussion centers on the transition from deterministic computing (traditional APIs/SaaS) to probabilistic computing driven by agentic systems, and the challenges and opportunities this presents for enterprise integration and IT operations.
2. Key Technical Insights
- Agentic Abstraction Layer: The shift from cloud computing to agentic computing represents a higher level of abstraction, aiming to offload human toil (decision-making, data parsing, code generation) to autonomous agents.
- Four Phases of Agent Collaboration: The IoA platform must address four critical stages: Discovery & Identity (finding agents and verifying trust), Composability (stitching agents into workflows), Deployment (scaling the workflow), and Evaluation (measuring the probabilistic outcomes and governance).
- Semantic vs. Syntactic Communication: A major technical gap is the lack of standards for the semantic layer (defining what terms like “best” or “90% confidence” mean across different vendor agents), which is more critical than the current focus on syntactic communication protocols (like A2A, MCP, ACP).
3. Business/Investment Angle
- Cisco’s Strategic Entry Point: Outshift positions Cisco to enter high-risk, emerging markets like agent platforms, reducing overall organizational risk by incubating these ventures internally before broader adoption.
- Eliminating IT Toil: Agentic systems promise massive efficiency gains. The “SRE Jarvis” example showed eliminating the work of three full-time engineers in support and reducing the time for complex configuration tasks (e.g., setting up new LLMs or EC2 instances) from days to minutes for 30% of tasks.
- Market Fragmentation Challenge: As major SaaS providers (Salesforce, Workday, ServiceNow) become agentic, the enterprise faces the complex problem of stitching together disparate, vendor-specific agents that speak different “languages” (frameworks and semantics).
4. Notable Companies/People
- Vijoy Pandey (Cisco Outshift): The central voice, outlining the vision for the Internet of Agents and the role of Outshift as an internal incubator for Cisco.
- Cisco: Positioning itself as a key player, leveraging its networking expertise to handle the high-speed, multimodal state exchanges required by real-time agent collaboration.
- Agency: Mentioned as the open-source manifestation/project designed to make the Internet of Agents concepts concrete.
- SRE Jarvis: A concrete internal multi-agent system built by Outshift to automate Site Reliability Engineering tasks, accessing tools like Kubernetes, PagerDuty, and AWS.
5. Future Implications
The industry is moving toward a future where complex business processes are solved by ensembles of specialized agents, mirroring human SME teams. This requires a foundational, standardized platform (the IoA) to manage discovery, trust, orchestration, and evaluation across vendor boundaries. The pain of this transition will mirror the early days of cloud adoption, requiring organizations to adapt to probabilistic workflows.
6. Target Audience
This episode is highly valuable for Enterprise Architects, AI/ML Strategy Leaders, CTOs, and Product Leaders focused on operationalizing AI, managing complex IT integration, and understanding the next evolution of enterprise software infrastructure beyond current LLM deployments.
🏢 Companies Mentioned
đź’¬ Key Insights
"And so, that's where this protocol bridges the agent layer to the actual infrastructure layer because, yes, you can do the framework or API-level communication here at the agent layer, but how do you translate it to the performance, SLO, efficiency, security guarantees that you need from the network?"
"So, everybody's journey is too early where you're saying, 'Let me handle the app-level problem, the agent-level problem,' but you're not looking at SLO guarantees, you're not looking at performance, you're not looking at efficiency, you're not looking at how it's got all... what to play out on the infrastructure that sits below."
"They need to provide the SLOs that nobody is talking about right now."
"But in the end, they need to sit, and the rubber needs to hit the road. So, they need to sit on the cloud-native and front. They need to sit on actual networking gear. They need to provide the SLOs that nobody is talking about right now."
"The goal that we all have is the agent itself—there's a supervisor agent that itself figures out there's a set of agents, there's how I structure it for a business workflow, and I'm just going to structure it or compose it on the fly. So, self-forming agents."
"let's make sure that security is quantum-safe right from the get-go because quantum is not that far off, and especially breaking security in the quantum domain is not that far off."