At Outshift by Cisco, we’re redefining platform engineering with the integration of agentic AI—an idea inspired by the vision of highly capable, autonomous systems that amplify human ingenuity. 

The role of platform engineering has grown increasingly multifaceted in the past decade. From the rise of Kubernetes and containerization to the explosion of cloud-native architecture, engineers are managing a vast ecosystem of intricate tools and technologies. The shift toward microservices has multiplied workloads and introduced new challenges, making cognitive overload and efficiency critical concerns.

Rethinking platform engineering with AI

We approach platform engineering with a forward-thinking perspective at Outshift. We see a future where AI plays an integral role in simplifying workflows and automating tasks. 

• Simplified learning: AI assistance helps engineers navigate the diverse cloud-native landscape without needing deep expertise in every technology, allowing new team members to learn faster.
• Self-service with a personal touch: Incorporating LLM (Large Language Model) reasoning into self-service features improves user experience and accessibility.
• Improved productivity: AI agents efficiently handle user queries by accessing knowledge bases, streamlining processes for both platform teams and users.
• Fostering innovation: Automating routine tasks with AI frees engineers to focus on creative projects and collaboration, enhancing engagement in higher-order work.

Meet JARVIS: Outshift’s AI Platform Engineer

At the heart of our AI initiatives is JARVIS, the persona behind a multi-agentic system currently with over 15 sub-agents, more than 40 tool calling agents and upwards of 10 self-service workflows. And yes, as you correctly guessed, JARVIS is inspired by Iron Man.

When we started this journey in April of 2024, the initial idea was too far-fetched. So, I was like, “Remember Tony Stark and JARVIS in Iron Man? Can we create a modern cloud infrastructure just like that?" 

From here, across several work streams, including three internship projects, were all combined to create JARVIS, the AI Platform Engineer as we know it today.

Key features of JARVIS

1. Knowledge management: JARVIS integrates with knowledge bases like docs, policies, code, Jira, and public expert knowledge using GraphRAG and LLMs to quickly derive insights from scattered data.
2. Self-service capabilities: Through multi-agent LangGraph, it proves self-service features, supporting tasks like Jira interactions and platform CI/CD bootstrapping for development and production on Kubernetes and VMs.
3. Code generation: It can generate Kubernetes configurations using a hybrid machine learning (ML) approach with LLMs and symbolic AI, making Kubernetes more accessible through natural language and diagrams instead of complex YAML configurations.

User interfaces of JARVIS

Recognizing the significance of seamlessly integrating with existing user workflows, we have developed JARVIS to effortlessly accommodate multiple user interfaces. Currently, Outshift users are utilizing the following four interfaces.

1. Backstage: We have an integrated chat assistant in our internal developer portal. Users prefer it over Backstage search or templates for workflow executions.
2. Webex: In addition to user interactions on instant messaging, this user interface is also useful as an effective notification channel including any secure information on top of Webex end-to-end encryption.
3. JIRA: As an augmented member of our team, JARVIS can fully handle certain JIRA tasks, including communicating with the reporter to obtain any missing information.
4. CLI: In addition to the functionality related to building and pushing devTest container images, this provides developers with all the capabilities of JARVIS at the shell.

Game-Changing K8s Dev Experience @ Outshift

Along the lines of more advanced AI capabilities, currently, our engineers are enjoying an innovative agent-driven experience in our EKS K8s sandbox. This setup allows fast natural language iteration cycles for deploying and troubleshooting apps.

As a developer, you can simply talk to JARVIS to deploy your container. JARVIS will generate all the required K8s configuration using a hybrid ML approach. JARVIS also has multiple sub-agents to handle tasks related to git, ECR and kubectl on behalf of the developer.

Furthermore, we are currently looking to get third party agents such as Komodor’s KlaudiaAI to directly collaborate with JARVIS leveraging distributed agent-to-agent communication.

Learnings in agentic AI so far

1. LLM reasoning is good, but you will realize the true potential of AI when you start assembling multi-agent systems to accomplish significantly more complex tasks. See our example of how agent-based systems can fully automate DevOps tasks.
2. There are many challenges and considerations to be made around AI, ethics, reliability, and team readiness. They all play critical roles in determining impact. For enterprises, an internal use case such as this, is a great way to rapidly iterate on AI’s potential and applications.
3. It is important to see how you can seamlessly integrate AI capabilities into existing user interfaces and workflows. We have developed it, so JARVIS feels like another team member working alongside us.

Creating the future of agentic AI in platform engineering

We are only at the beginning in exploring the intersection of agentic AI and platform engineering. Our goal is to enable teams to seamlessly integrate with agentic systems that amplify their potential, encourage collaboration, and inspire innovation. We’re not just building AI agents; we’re redefining the potential for platform engineering teams globally. 

Explore more about how Outshift is building an open, secure, and scalable infrastructure to support the future of agentic AI and stay tuned for more platform engineering initiatives.