Article originally posted on InfoQ. Visit InfoQ

Cisco’s innovation arm, Outshift, has unveiled JARVIS. An AI-powered assistant designed to streamline platform-engineering workflows. JARVIS offers a conversational interface that simplifies complex tasks, reducing both execution time and cognitive overhead.

Cisco has developed JARVIS as a member of the platform-engineering team. JARVIS can integrate with over 40 tools and, among other tasks, can seamlessly provision infrastructure, onboard new applications to CI and retrieve important documents with all instructions originating from natural language. The team at Cisco communicate with JARVIS using tools they already use every day, such as Jira, Webex and Backstage. The team can assign Jira tickets to JARVIS, and it executes the work.

The results have been significant. In a LangChain blog post, they reported that “Tasks that previously took a week, such as setting up CI/CD pipelines, can now be completed in under an hour.” According to a  NetworkWorld blog post, “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.”/p>

JARVIS is powered by a hybrid AI architecture that combines multiple techniques. Built using LangGraph, the main engine behind JARVIS is large language models (LLMs) that provide the natural-language processing capability to best interpret developer requests. However, the responses generated by the LLMs are not immediately trusted. JARVIS implements three additional safeguards to ensure accuracy and reliability. First, it uses rule-based validation through symbolic logic to confirm that outputs adhere to predefined standards, such as naming conventions. Second, it enforces clearly defined, repeatable workflows to ensure tasks are executed consistently, without allowing the AI to improvise. Finally, it introduces agent supervision, where one agent reviews and verifies the work of another, adding an additional layer of quality control.

To answer knowledge questions, JARVIS leverages retrieval-augmented generation (RAG) by allowing its AI agents to supplement their in-memory knowledge by looking up information from external sources of structured data like wikis and codebases.

At present, JARVIS is only used internally at Cisco. However, the company has announced plans to open-source key components such as its integration with Backstage and standalone agents. The project is also contributing to the Internet of Agents initiative of intelligent software agents that can communicate and collaborate across shared protocols.

Bill Gates, co-founder of Microsoft, sees AI agents like JARVIS as a part of the future. In a 2024 blog post, he noted, “Agents are not only going to change how everyone interacts with computers. They’re also going to upend the software industry, bringing about the biggest revolution in computing since we went from typing commands to tapping on icons.”