Published on 00/00/0000
Last updated on 00/00/0000
Published on 00/00/0000
Last updated on 00/00/0000
Share
INSIGHTS
5 min read
Published on 01/30/2025
Last updated on 01/30/2025
Agentic AI for troubleshooting and change management
Share
At Outshift, we are addressing long-standing challenges in IT operations by leveraging generative, agentic AI. Despite many technological advances, cross-domain and cross-layer troubleshooting, root-cause analysis, and change management are still largely manual tasks. Why is this the case?
The limitations of traditional IT troubleshooting approaches
You know the situation well: The network department has updated the firewall software and tweaked the ruleset. The team has tested the update and everything seems to be working fine. The next day, the application department gets an alert that the backup of the application database did not complete last night, but from peeking at their observability system they have no idea what went wrong.
By their very nature, cross-domain issues span multiple organization teams — network, application, database, and more. Each team’s tools and interpretations of the data differ, creating silos of information, making it difficult to connect the dots. This is a prime example of Conway's Law at work, where the structure of the software mirrors the structure of the organization.
The classic approach to overcoming organizational boundaries and solving a cross-domain problem or creating a plan for a change that affects multiple domains, involves gathering stakeholders from all departments to compare their observations and notes to arrive at a solution approach. While this yields results, it is a costly, slow, and inefficient process.
Eliminating the need for the meeting has been our goal ever since. A common approach often suggested is to aggregate all the data from the different organizations into a unified schema and create a common interpretation across departments. While feasible for some this approach has proven challenging for large and diverse organizations. Microsoft CEO Satya Nadella summed up the challenges nicely in the recent BG2 podcast: "Schematizing the world [...] is simply impossible to do.”
A new approach in IT troubleshooting with agentic systems
Enter the era of Large Language Models (LLMs) and agentic systems, which offer an alternative to schema-dependent reasoning. Instead of reasoning about data using well-defined schemas and decision logic, what if we approach cross-domain reasoning like humans do in a meeting, as a collaborative discussion using natural language? Agentic AI allows us to replace the human meeting with a "meeting of specialized agents." These agents can solve a problem or plan a change in an iterative way like humans.
Agents quickly hypothesize what might have caused or could cause a problem, validate or disprove a hypothesis using data retrieved from the various observability solutions, knowledge bases (like Cisco's 40 years of collected data on problems and their solutions), the Internet, and adjust the hypothesis using the retrieved data in an iterative manner to arrive at a conclusion. Agents not only include postmortem forensic data, but also interface directly with live IT systems, mimicking the way humans troubleshoot. Agents are not limited to post-deployment analysis but can create models (think digital twins) of a deployment to proactively address issues, minimize disruptions, reduce risk, and optimize performance.
The key advantage of agent-based systems is their ability to perform these tasks quickly and efficiently, eliminating the need for traditional meetings and the complexities of data schematization. By automatically incorporating first-level investigations at ticket creation through platforms such as ServiceNow or ThousandEyes, organizations can quickly resolve cross-domain issues and promote a proactive approach to IT management.
Agentic collective memory
The effectiveness of agent-based systems depends on the concept of collective memory and learning. Individual agents have the ability to memorize and learn from past experiences, drawing conclusions quickly and without repetitive loops. For our diagnostic system, retrieval agents, hypothesis agents, validation agents, model building agents, etc. work together. The collective memory of the agents, formed through their interaction and their results is an essential component of agentic systems for pre- and post-deployment diagnostics. Agents learn as a group. They build a comprehensive knowledge base, similar to organizational learning.
Ongoing memory formation is critical in determining which information is retained and which is discarded. As they learn, agentic systems must exercise discrimination, recognizing that what is unimportant today may become important tomorrow. For example, troubleshooting a security issue in a lab environment might be disregarded because it is “just the lab” but becomes essential when transitioning to production.
Explore the future of IT management with agentic AI
As we further explore the capabilities of agentic systems, these advancements hold potential for transforming IT troubleshooting and management. By shifting services to software, organizations can achieve efficiency, reduce downtime, and enhance operational performance.
If you're curious to learn more about this technology and its applications, join me at Cisco Live in Amsterdam Feb. 9-14, 2025. I’m hosting four sessions which explore the intricacies of agentic systems, covering everything from pre- and post-deployment planning to troubleshooting and change management:
- AIHUB-2170 - AI Agents and Agentic Frameworks: An Overview
- BRKETI-2172 - A New Era of Cross-domain Diagnostics with AI and Agentic Systems
- DEVNET-2094 - Agentic AI for Troubleshooting and Change Management: A Glimpse at the Future
- AIHUB-2006 - Applying AI/ML Methods to Diagnose Network Issues Using Telemetry Data
See you in Amsterdam!
Excited to learn more about what the Outshift team is up to? Stay tuned for more TechShifts, our new video and blog series where Outshift that gives you a front-row seat to cutting-edge work shaping the future of tech. In this series, our Distinguished Engineers, Researchers, and Product Managers share exclusive insights into the groundbreaking projects they're working on right now.
Subscribe to our YouTube channel to hear directly from these experts who are pushing the boundaries and keeping us at the forefront of AI, quantum, next-gen infra and more.
Subscribe to
the Shift!
Get emerging insights on innovative technology straight to your inbox.
Welcome to the future of agentic AI: The Internet of Agents
Outshift is leading the way in building an open, interoperable, agent-first, quantum-safe infrastructure for the future of artificial intelligence.
