Uploaded May 2026 | Updated September 2026, 1 week ago
Ken Duda, CTO, showcases Arista's innovative approach to network management using artificial intelligence. The presentation highlights One AVA, a standalone application that integrates multiple tools within the Arista suite, such as Cloud Vision, CVQ, AGNI, VeloCloud, and NDR, to provide a unified troubleshooting experience. Powered by large language models (LLMs) and advanced techniques like Retrieval-Augmented Generation (RAG), AVA operates autonomously by constantly analyzing sessions, extracting learnings, and managing customer-specific memory databases within the secure Network Data Lake (NetDL). This architecture helps bridge the gap between deterministic software logic and probabilistic LLM behaviors, enabling operators to streamline troubleshooting workflows.
During the demonstration, the speakers walk through several realistic scenarios, including how a trouble ticket initiated by a webhook triggers an autonomous investigation by AVA. The assistant leverages skill documents, performs endpoint searches, checks Wi-Fi connectivity records, and even runs continuous reachability tests to isolate issues, such as a misconfigured firewall rule or a lack of RADSec connection. The human-in-the-loop design ensures that any corrective adjustments, like updating ACL policies or deploying access point configurations, require explicit administrator approval before execution. This process is supplemented by synthetic client connectivity tests run by the access point's multi-function fourth radio, allowing AVA to validate end-to-end network resolution directly from the client's perspective.
The presentation addresses practical enterprise concerns, particularly regarding security, data privacy, and the inherent unpredictability of LLMs. To maintain strict data isolation, AVA isolates each customer's learnings and historical context within their specific tenancy, using human engineers rather than automated AI to generalize new skills for the broader knowledge base. Additionally, the platform mitigates risks from malicious data poisoning and prompt injection by strictly enforcing role-based access control (RBAC) and routing all structural network changes through formal change-control processes rather than executing them blindly. Ultimately, Arista positions AVA not as a replacement for human engineering, but as an essential, proactive assistant that turns reactive network troubleshooting into a highly efficient, data-rich operational workflow.
Presented by Ken Duda, CTO, and Sriram Venkiteswaran, Senior Director of Product Management. Recorded live at Mobility Field Day 14 in San Jose, CA on May 8, 2026. Watch the entire presentation at techfieldday.com/appearance/arista-presents-at-mobility-field-day-14 or visit TechFieldDay.com/event/mfd14 or Arista.com for more information.
Ken Duda, CTO, showcases Arista's innovative approach to network management using artificial intelligence. The presentation highlights One AVA, a standalone application that integrates multiple tools within the Arista suite, such as Cloud Vision, CVQ, AGNI, VeloCloud, and NDR, to provide a unified troubleshooting experience. Powered by large language models (LLMs) and advanced techniques like Retrieval-Augmented Generation (RAG), AVA operates autonomously by constantly analyzing sessions, extracting learnings, and managing customer-specific memory databases within the secure Network Data Lake (NetDL). This architecture helps bridge the gap between deterministic software logic and probabilistic LLM behaviors, enabling operators to streamline troubleshooting workflows.
During the demonstration, the speakers walk through several realistic scenarios, including how a trouble ticket initiated by a webhook triggers an autonomous investigation by AVA. The assistant leverages skill documents, performs endpoint searches, checks Wi-Fi connectivity records, and even runs continuous reachability tests to isolate issues, such as a misconfigured firewall rule or a lack of RADSec connection. The human-in-the-loop design ensures that any corrective adjustments, like updating ACL policies or deploying access point configurations, require explicit administrator approval before execution. This process is supplemented by synthetic client connectivity tests run by the access point's multi-function fourth radio, allowing AVA to validate end-to-end network resolution directly from the client's perspective.
The presentation addresses practical enterprise concerns, particularly regarding security, data privacy, and the inherent unpredictability of LLMs. To maintain strict data isolation, AVA isolates each customer's learnings and historical context within their specific tenancy, using human engineers rather than automated AI to generalize new skills for the broader knowledge base. Additionally, the platform mitigates risks from malicious data poisoning and prompt injection by strictly enforcing role-based access control (RBAC) and routing all structural network changes through formal change-control processes rather than executing them blindly. Ultimately, Arista positions AVA not as a replacement for human engineering, but as an essential, proactive assistant that turns reactive network troubleshooting into a highly efficient, data-rich operational workflow.
Presented by Ken Duda, CTO, and Sriram Venkiteswaran, Senior Director of Product Management. Recorded live at Mobility Field Day 14 in San Jose, CA on May 8, 2026. Watch the entire presentation at techfieldday.com/appearance/arista-presents-at-mobility-field-day-14 or visit TechFieldDay.com/event/mfd14 or Arista.com for more information.
