When it comes to developing highly intelligent AI agents, one might not think of combining open systems technology and the theoretical super-logic behind a movie character, but Cisco’s Outshift development team is doing just that.
The Outshift by Cisco team is developing JARVIS, an AI platform engineer and the persona behind a multi-agentic system that includes more than 15 specialized sub-agents, 40 tool integrations, and 10 automated workflows that work together to handle complex engineering tasks, says Hasith Kalpage, CISO and platform engineering director at Outshift by Cisco.
The aim is to help address the growing complexity of platform and infrastructure engineering, and the inspiration is from the “Iron Man” superhero franchise: Tony Stark’s superlative natural-language computer system is called Just A Rather Very Intelligent System (J.A.R.V.I.S.).
JARVIS sub-agents include a Kubernetes code-generation agent that translates natural language into K8s configurations; a knowledge management agent using GraphRAG to integrate with documentation; an agent for repository operations; and a container registry agent for image management.
JARVIS tools include AWS services such as EC2, EKS and S3, Kubernetes operations, and JIRA for ticket handling and user communication, Kalpage said. Workflow support includes natural language Kubernetes and infrastructure provisioning without deep technical knowledge.
“When we initiated the project, the concept and its potential seemed quite ambitious, akin to Tony Stark creating his Iron Man suit, with JARVIS used as a reference point for tackling similar challenges in the complex cloud-native environment,” Kalpage said.
“We deliberately named it after Iron Man’s AI assistant because we wanted that level of capability – an intelligent system that understands context, can access different tools and knowledge bases, and most importantly, works alongside engineers rather than just responding to commands,” Kalpage said.
Platform engineering has hit a crisis point of complexity. The modern tech stack of Kubernetes, microservices, and cloud-native tools has created three critical pain points, Kalpage said, including:
- The bottleneck problem: Engineers spend up to 70% of their time on repetitive tasks rather than innovation. JARVIS automates these workflows, turning day-long processes into minute-long ones.
- The expertise gap: No human can master every component in today’s cloud stack. JARVIS bridges this gap by encoding platform knowledge and providing contextual assistance when engineers need it.
- The integration nightmare: Most enterprises have dozens of disjointed tools. JARVIS creates a unified interface across these systems, orchestrating complex workflows that would normally require manual coordination across multiple platforms.
Technically, JARVIS operates as a distributed brain with four interfaces, including a Backstage portal, Webex, JIRA, and CLI, meeting engineers in their existing workflows, Kalpage said. “Under the hood, it uses a LangGraph architecture with supervised, specialized, and reflection agents working together in feedback loops.”
“It connects to knowledge bases, executes tool calls to various systems, and even generates Kubernetes configurations using hybrid ML approaches,” Kalpage said.
The JARVIS architecture aligns with Cisco Outshift’s Internet of Agents and recently announced AGNTCY (pronounced “agency”) initiative.
Outshift describes the Internet of Agents as standards-based, shared infrastructure components that enable quantum-safe, agent-to-agent communications. When AI agents begin to proliferate, a new, open structure will be needed so they can securely communicate and collaborate together to solve complex problems, Cisco stated. The “Internet of Agents” is an open-sourced, three-layer architecture that would enable quantum-safe, agent-to-agent communication to allow AI agents to collaborate autonomously and share complex reasoning.
The Internet of Agents will connect AI systems across vendor and organizational boundaries, and technical frameworks. “Without this foundation, we’re essentially trying to build the web without RPCs, HTTP, DNS, or TCP/IP,” according to a recent blog by Vijoy Pandey, senior vice president of Outshift by Cisco.
AGNTCY, meanwhile, will help develop the industry open-source cooperation Cisco Outshift envisions for the Internet of Agents and more.
“Without standardized infrastructure sourced from high fidelity open-sourced options, builds are happening in silos and limiting what’s possible with multi-agent systems,” Pandey stated. “Our collective approach breaks down these barriers by establishing open specs and reference implementations with architecture built with open-source code that tackles the end-to-end requirements for sourcing, creating, scaling, and optimizing agentic workflows from a single app to thousands,” Pandey stated.
As of its March introduction, AGNTCY includes technology from Cisco, LangChain with its agent orchestration technology, and Galileo with security and observability capabilities.
“Multi-agent systems are the future of AI, but we need open standards for both collaboration and rigorous assessment. The AGNTCY creates the foundation for a truly observable and trustworthy AI ecosystem. Our evaluation frameworks ensure agents perform reliably across diverse scenarios, empowering organizations to deploy at scale,” stated Harrison Chase, CEO of LangChain, in Pandey’s blog.
“At Galileo, we help organizations build trustworthy agents and scale agentic systems with confidence,” stated Yash Sheth, co-founder of Galileo.
The evolution path for JARVIS directly aligns AGNTCY. “We’re converting it from a monolithic architecture to leverage the Internet of Agents framework and protocols. This means breaking it into distributed components that communicate through standard interfaces like Agent Connect Protocol and Agent Gateway Protocol – the core standards we’re developing with AGNTCY,” Kalpage said.
JARVIS serves as our real-world proof point for Internet of Agents – demonstrating both the need for standardized agent communication and the practical benefits when implemented properly, Kalpage said.
As for whether or not JARVIS will be a Cisco product at some point soon, Kalpage said the company is approaching JARVIS as an open-source initiative rather than a traditional product.
“We’re collaborating with Cloud Native Computing Foundation organizations and planning to open-source key components starting with our Backstage chat assistant and several standalone agents,” Kalpage said.
Beyond the technology, Iron Man’s J.A.R.V.I.S. is known for his sarcastic and snarky demeanor, which sadly sounds like it will be missing from any productized version of the system.
“Our JARVIS stays professional for now, though engineers occasionally request a ‘sarcasm slider’ to dial up the personality when needed!” Kalpage said. “We’ve prioritized making JARVIS a reliable team member rather than giving it Hollywood-style snark.”
Source:: Network World