Detailed write-ups
ONUG AI Networking Summit Dallas (May 13–14)
ONUG’s spring summit at the Hyatt Regency Frisco brought IT leaders together around five tracks — AI Infrastructure, AI Networking, AI Security, AI Automation, and Special Programs. Cisco’s on-stage focus was “From AI Ambition to AI Infrastructure” (Sai Natarajan) and “Hybrid AI Infrastructure and AgenticOps Security in Practice.” Craig Connors (VP, Cisco Security Business Group) led a panel on Defending Against Adversarial AI Agents. Cisco highlighted Cisco Nexus One as its agentic-era data-center anchor. ONUG remains the most concentrated venue for cross-vendor agentic-NetOps conversation.
Sources: ONUG · Cisco Blogs
HPE: self-driving networks for Mist + Aruba Central
HPE positions itself as the “industry’s first and only fully autonomous, agentic AIOps networking” provider. The May 6 launch shipped autonomous actions across HPE Mist and HPE Aruba Central — capacity and radio optimization, self-securing actions, user roaming issue resolution — backed by a microservices architecture and an “advanced agentic mesh.” First AI-native dual-platform Wi-Fi APs (HPE Networking 723H) are GA, including wall-mount form factors for hospitality.
Sources: HPE · Network World · The Register
Extreme Networks Agent ONE (Extreme Connect 2026)
Second-gen AI agent package announced May 5. Two operating modes: Agent ONE Coworker (July 2026 GA) sits alongside IT teams with proactive, context-aware suggestions and machine-speed automated execution; Agent ONE Operator (Q4 2026 GA) is always-on and autonomous within governance boundaries. Extreme Exchange debuts as an “AI skills marketplace” for industry-specific extensions. Forrester’s post-event read: momentum hinges on Platform ONE adoption and how cleanly the AI capabilities differentiate from cloud-NOC peers.
Sources: Extreme Networks · SiliconANGLE · Forrester · Help Net Security
Cisco Crosswork & AgenticOps (foundational)
Cisco’s vision is NOCless / White NOC / Dark NOC — reducing or eliminating manual intervention as the operating model. The Crosswork Multi-Agentic AI Framework is containerized, supports both LLM- and ML-based agents, and orchestrates agent-to-agent communication. Specialized agents: configuration drift, toxic-factor detection, net-query, performance reporting, troubleshooting. A knowledge graph semantically aligns all agents through a digital network representation. Built on the TM Forum Incident Co-Pilot Catalyst project.
Sources: Cisco Crosswork Multi-Agentic AI · Cisco AgenticOps
NetBrain R12.3 — ReAct and AI Deep Diagnosis (foundational)
First major release under new CEO Bernadette Nixon. Headline is AI Deep Diagnosis, which uses a ReAct (Reasoning + Acting) framework to iteratively query NetBrain’s automation library, working off NetBrain’s digital twin and live network data. The agent renders its reasoning on a network map so engineers can follow how it got to a conclusion. R12.3 also ships an AI Runbook Companion and AI Ticket Analysis. In testing, the system handled 90% of real-world network issues. Nixon’s framing: “agentic NetOps” — AI acting alongside engineers rather than just assisting them.
Sources: Network World · NetBrain · Channel Insider
NetBox Labs — source-of-truth-grounded copilot (foundational)
NetBox Copilot is “enterprise-ready AI grounded in accurate infrastructure data.” Natural-language interface for infrastructure investigation, change validation, dependency assessment before maintenance windows. April 2026 partnership with AHEAD embeds the platform into AHEAD’s networking + cloud + AI infrastructure stack. The strategic positioning: system-of-record-to-system-of-control, making engineers “masters of intent.”
Sources: Network World · NetBox Labs
Aviz Networks Network Copilot 1.5.6 (foundational)
Vendor-neutral, AI-native network observability + operations assistant. BYO-LLM with a Python SDK for customization. R1.5.6 introduced “agentic NOC” workflows and custom agent uploads. Aviz’s pitch is “AI NOC, not AI-on-NOC” — flipping the stack so AI is the operations primitive rather than an overlay on a legacy NOC.
Sources: Aviz Networks
IBM Network Intelligence (Think 2026)
Network-native AI built on the human + AI partnership model. Pairs IBM Granite Time Series Foundation Models (pre-trained on telemetry, alarms, flow data) with LLM-powered reasoning agents that hypothesize root causes and filter noise to deliver high-confidence insights. Built in collaboration with IBM Research.
Sources: IBM Newsroom
NVIDIA BlueField-4 — the agentic-AI DPU substrate
Announced at GTC 2026 (March 16). 800 Gbps throughput, 6× compute over BlueField-3, 64 Arm cores. The new BlueField-4 STX storage architecture targets the data-access bottleneck for agentic AI inference. Cisco has extended its Hybrid Mesh Firewall into the BlueField ecosystem — partnering with AT&T to push these AI-factory concepts into mobility networks. Agentic-NetOps implication: more network functions move out of CPUs into the DPU, freeing the host for AI workloads.
Sources: NVIDIA Blog · Tom’s Hardware
Microsoft NiC + AIOps (internal blueprint)
Microsoft’s Network Infrastructure Copilot (NiC) — natural-language interface for network practitioners across network health, configuration state, documentation, troubleshooting, and live device data. Paired with an AIOps automation layer that prevents and resolves issues before they impact users. Reported impact: thousands of engineering hours saved and a significant reduction in disruption time across Microsoft’s global workforce. Useful template for enterprise-internal agentic-NetOps build vs. buy.
Sources: Microsoft Inside Track
The Fast Mode — “the year of autonomous networks”
Industry-wide framing: 2026 marks the transition from agentic-NetOps pilots to production-scale deployments delivering quantifiable ROI. The publication’s recurring themes — shared ontologies, interoperable APIs, telco knowledge graphs, planner + verifier architecture, accountable intelligence — line up with what vendors are shipping. No single telco, vendor, or hyperscaler can deliver fully autonomous networks alone; partnership-driven, open models are winning.
Sources: The Fast Mode
|