2026 Showcase

2026 AI for Networking


Welcome to the NextGenInfra AI for Networking Showcase, our research guide for enterprise and carrier network executives navigating the shift from AIOps to agentic network operations.

This year's report, AI for Networking: Charting the Agentic Waters, looks at why context, governance, and trust define the next wave of autonomy — how reasoning agents move operations from detecting problems to explaining and resolving them, and what it takes to let them act safely in production networks. Download the report by clicking the button on the right (ungated - no personal information required). New to the series? Our 2025 edition covers the first wave of agentic frameworks, and this year's report grades how its predictions held up.

Thanks to our sponsors for making this resource possible!

Courtesy of our sponsors!

Complimentary Report Download

AI for Networking: Charting the Agentic Waters

No personal info needed to download

Download "AI for Networking: Charting the Agentic Waters"

By downloading this report, you agree to our Terms of Use and Privacy Policy.

Subscribe to our newsletter and receive new report updates

The monthly AvidThink infrastructure brief. No spam — unsubscribe anytime.

Analysis by AvidThink

What the 2026 report finds

Two years into a five-to-seven-year journey to full autonomy, agentic AI is doing real work in network operations — but in a far more constrained way than 2025's announcements implied. Across the report and our sponsor interviews, five themes define where agentic networking stands in 2026.

  1. The harness, not just the model, made agents useful

    Agents now investigate, correlate, and recommend across enterprise, carrier, and data center networks. The leap came from control patterns borrowed from coding agents — curated context, action governance, a deterministic execution layer, and token budgets — more so than innovations in foundation models. Per-token prices fell but bills rose, so tokens are now budgeted like headcount.

  2. Context is the new competitive moat

    An agent is only as accurate as the source of truth, ontology, and digital twin beneath it. Who owns that context layer is 2026's buy-versus-build fault line: tier-1 telcos treat their ontologies as intellectual property, while vendors pitch theirs as product differentiators.

  3. Autonomy is granted one action class at a time

    Generally available vendor products stop at pre-authorized action classes — HPE's trusted actions, Extreme's Coworker mode. Only a handful of operator-built carrier loops run unattended, each for one scenario in one domain.

  4. Governance shifts from approvals to evidence

    Agent sprawl forced a governance plane: agent identity, registries and gateways, policy-as-code, autonomy settings, and an immutable audit trail. Vendors argue approval-based governance cannot keep pace with AI-driven change, and agentic platforms move operators from human in the loop to human on the loop by providing reasoning logs and auditable runbooks.

  5. Agents reason; deterministic automation acts

    Changes reach the network only through proven workflows, controllers, and GitOps pipelines — elements of pre-agentic network automation. This year's best published automation outcomes came from deterministic automation and a clean source of truth, not from agents alone.

Download the full report →

Highlights from industry thought leaders

Director’s Cut

Extended, in-depth versions of select interviews.

Director’s Cut
Blue Planet (a Ciena Company)

Governing Autonomous Networks

  • Human-centric, approval-based governance cannot keep pace with the scale and impact of AI-driven decisions, so CSPs need an evidence-based approach.
  • Operators must understand the consequences of overlapping AI actions — blast radius, how far to roll back, and whether to correct course going forward.
  • Blue Planet config change management immutably records who initiated an event, why, the context used, the AI's decisions, and the devices affected.
Abstract
Balaji Subramaniam, Director, Product Management, Blue Planet at Ciena, argues that human-centric, approval-based governance cannot keep pace with the scale and impact of AI-driven decisions in autonomous networks, so communications service providers need an evidence-based approach that reveals the consequences of overlapping AI actions, including blast radius and rollback scope. He explains how Ciena's config change management immutably captures who initiated an event, why, the context used, the AI's decisions, and the affected devices, embedding governance directly into the network for audits and reporting.
Director’s Cut
Extreme Networks

Agentic AI Ends Firefighting With Trusted Autonomy

  • Extreme Agent ONE, embedded in Extreme Platform ONE, runs in Coworker mode to reason and recommend and in Operator mode to execute changes autonomously.
  • Extreme favors "trusted autonomy" over unrestricted AI access, with governance, context, and safety controls growing in importance as autonomy rises.
  • New Perspective Senior Living uses AI-based templates with its own guardrails to launch nine new communities in a week instead of months.
Abstract
Monica Kumar, EVP & CMO of Extreme Networks, explains how agentic AI moves network operations from reactive firefighting to proactive management, with Extreme Platform ONE and its embedded Extreme Agent ONE detecting, explaining, and resolving issues in Coworker and Operator modes under a model of "trusted autonomy." She cites New Perspective Senior Living, which uses AI-based templates with its own guardrails to launch nine new communities in a week, and argues that success depends on the platform, data, and context behind the agent rather than on simply having a chatbot.
Director’s Cut
Gluware

AI Threats Make Network Automation a Top Priority

  • AI-driven threats, accelerated by frontier models such as Mythos, are making network automation a top priority for CTOs and infrastructure leaders.
  • With up to 30% of configurations bespoke and depreciation stretching to seven or eight years, ripping and replacing equipment is impractical.
  • Gluware's proof-of-trust architecture grounds AI in the customer's own configurations, and Titan Exposure Management pinpoints only the devices that are actually vulnerable.
Abstract
Ernest Lefner, Chief Product Officer at Gluware, argues that AI-driven threats accelerated by frontier models like Mythos are making network automation a top priority, especially since "automation resistant" enterprise networks with bespoke configurations and stretched depreciation cycles make ripping and replacing equipment impractical. He describes Gluware's "proof of trust" architecture, Titan Exposure Management for pinpointing vulnerable devices, and its vendor-agnostic DIAL layer, while recommending a crawl-walk-run approach that lets enterprises rethink how they automate and secure their networks.
Director’s Cut
HPE

Self-Driving Networks Promise 10x Faster Deployments

  • HPE's AI-native, self-driving network is in production today, delivering deployments up to 10x faster through zero-ticket cutovers.
  • A large healthcare institution reports 90% fewer user complaints, and HPE challenges enterprises to test it on their worst site.
  • Costly cloud tokens are pushing inference to the edge, which HPE covers with one AI visibility layer across Wi-Fi, switching, SD-WAN, NAC, and the data center.
Abstract
In an interview at HPE Discover, Sudheer Matta, Senior Vice President, Products at HPE, explains how HPE's AI-native, self-driving network is in production today, delivering deployments up to 10 times faster through "zero ticket cutovers," up to 90% fewer user complaints, and greater workflow automation. He also describes how costly cloud tokens are pushing AI inference to the edge, a shift HPE supports with an end-to-end AI visibility layer spanning Wi-Fi, switching, SD-WAN, NAC, and the data center that frees IT staff from routine firefighting to focus on strategic work.
Director’s Cut
Nokia

Agentic AI for Network Operations

  • Nokia's NSP unifies network data through ETL, curation, and ontologies, then layers on an agent and context framework with memory, RBAC, and validation.
  • NSP supports large language models from Google, Anthropic, and OpenAI so customers can balance performance against token costs.
  • Purpose-built troubleshooting, assurance, and insight agents aim to move customers from TM Forum Level 2 to Levels 3 and 4 with lower risk.
Abstract
Mike Thompson, VP, Head of Network Automation Product Management at Nokia, outlines Nokia's agentic operations strategy for its Network Services Platform (NSP), which combines unified network data, an agent and context framework, support for multiple large language models, and purpose-built troubleshooting, assurance, and insight agents. He positions NSP as an open, multi-vendor platform spanning IP and optical layers that helps customers advance from TM Forum autonomous networking Level 2 to Levels 3 and 4 through lower-risk automation that improves service, reduces costs, and boosts productivity.
Director’s Cut
Ribbon, AWS

Agentic AI for Telecom Ops

  • Rising 5G and AI complexity, compressed margins, and retiring experts form a perfect storm for service provider operations.
  • Ribbon's Acumen agentic AIOps platform pairs domain intelligence with Amazon Neptune, Bedrock, and AgentCore to cut root-cause analysis from hours or days to minutes.
  • A crawl-walk-run approach keeps humans in control, starting with NOC assistance and change management, with closed-loop remediation expected within two to three years.
Abstract
Ram Ramanathan, VP Product Management – AI, Automation & Cloud at Ribbon, and Chris McKenna, Global Head of Business Development, Telecoms at Amazon Web Services, discuss how AI is reshaping service provider operations amid rising network complexity, compressed margins, and retiring experts, with Ribbon's Acumen agentic AIOps platform combining domain intelligence with Amazon Web Services tools such as Amazon Neptune, Amazon Bedrock, and AgentCore to cut root-cause analysis from hours or days to minutes. The speakers recommend a crawl-walk-run approach that keeps humans in control and builds in guardrails and full observability, starting with low-risk use cases such as NOC assistance and change management, and they expect intent-based automation and autonomous closed-loop remediation within two to three years.

Video interviews

BE Networks

AI Networks at 150,000-GPU Scale

  • BE Networks builds networks for AI — including a 150,000+ GPU deployment with IREN — and uses AI to run them better.
  • Avoiding congestion can require action within nanoseconds, far faster than human operators can respond.
  • Time series and graph databases work with AI agents to diagnose, predict, and prevent failures, which is essential for neoclouds at scale.
Abstract
Amir Elbaz, CEO of BE Networks, explains how his company both builds networks for AI, including a deployment with partner IREN bringing up more than 150,000 GPUs, and applies AI to improve network utilization and resiliency at speeds far beyond human response. Using time series and graph databases working with AI agents, BE Networks diagnoses issues and predicts and prevents failures before they occur, a capability Elbaz positions as essential for neocloud operators running AI infrastructure at scale.
Blue Planet (a Ciena Company)

When AI Runs the Network, Who Governs?

  • Governance of autonomous networks has become a recurring boardroom topic for communications service providers worldwide.
  • Approval-based governance gives way to an evidence-based approach that reconstructs and explains AI actions after they are taken.
  • When no human is available to approve actions, governance must be embedded in the network itself — in how it decides, reacts, and presents evidence.
Abstract
Balaji Subramaniam, Director, Product Management, Blue Planet at Ciena, argues that human-centric, approval-based governance cannot keep pace with the scale and impact of AI-driven decisions in autonomous networks, requiring an evidence-based approach that reconstructs and explains actions after the AI has taken them. He contends that operators must understand the consequences of continuous, overlapping AI decisions, including blast radius and rollback, and that governance must be embedded directly into the network so it can govern itself and present evidence of its actions.
Cisco Systems

AI Traffic Doubles Every 6 Months: Why Networks Need Agentic Operations

  • AI-driven traffic is doubling every six months across Cisco customer networks, and agent tasks generate up to 450% more traffic per task.
  • Agents shift operations from human in the loop to human on the loop, showing their work through reasoning logs and pre-approved, auditable runbooks.
  • Cisco Cloud Control, now generally available in the US, unifies networking, security, and observability data from ThousandEyes and Splunk as context for agentic ops.
Abstract
Joe Vaccaro, SVP Network Platform & Assurance at Cisco Systems, explains how agentic AI is reshaping both network operations and workloads, with AI-driven traffic doubling every six months and agent tasks generating up to 450% more traffic, while agents shift operations from "human in the loop" to "human on the loop" through reasoning logs and auditable runbooks. He recommends starting with repeatable, lower-risk events and positions the newly available Cisco Cloud Control, which unifies networking, security, and observability data from ThousandEyes and Splunk, as the platform where agentic ops comes to life.
Extreme Networks

Agentic AI for Networking: Highlights

  • Agentic networking changes how network teams spend their time rather than cutting headcount, as complexity grows while budgets and staffing stay flat.
  • AI for networking has moved from AIOps anomaly detection to plain-English generative AI queries to agents that explain, recommend, and act within guardrails.
  • The differentiator is what an agent connects to, the context it uses, and the governance around it — a platform question, not a chatbot feature.
Abstract
Monica Kumar, EVP & CMO of Extreme Networks, explains that agentic networking is about changing how network teams work rather than reducing headcount, tracing AI's evolution from AIOps anomaly detection to plain-English generative AI queries to agentic AI that explains issues, recommends next steps, and acts autonomously within guardrails. She highlights Extreme Platform ONE and its embedded Extreme Agent ONE, which runs in Coworker and Operator modes, and positions agentic AI as a platform question defined by connectivity, context, and governance that frees network professionals from firefighting so they can focus on strategic work.
Gluware

AI Threats Make Network Automation Urgent

  • AI-driven threats make network automation and agentic operations an urgent necessity rather than a nice-to-have.
  • Up to 30% of configurations are bespoke snowflakes, and even greenfield networks quickly drift and turn brown without automation.
  • Gluware's DIAL layer replaces sprawling Ansible and Python scripts with one validated, consistent automation mechanism.
Abstract
Ernest Lefner, Chief Product Officer at Gluware, argues that AI-driven threats, bespoke "snowflake" configurations, and the tendency of even greenfield networks to quickly "turn brown" make network automation and agentic operations an urgent necessity built on proper configuration standards and data modeling. He highlights the Gluware platform and its Device Interaction and Automation Layer (DIAL) as a trusted, validated alternative to sprawling self-built Ansible and Python scripts, letting organizations focus on intent and business outcomes.
HPE

Self-Driving Networks Cut Complaints 90%

  • HPE frames its self-driving network around zero-ticket outcomes, citing deployments 10x faster than previous vendors.
  • The world's largest healthcare institution reports 90% fewer complaints from doctors and staff after deployment.
  • The University of Texas at Dallas now runs its network with one engineer instead of four, freeing staff from firefighting.
Abstract
Sudheer Matta, Senior Vice President, Products at HPE, presents the company's self-driving network strategy built around "zero ticket" outcomes, citing deployments 10 times faster than previous vendors and a healthcare institution reporting 90% fewer complaints from doctors and staff. He emphasizes that trust and proven results matter more than marketing, challenging enterprises to test HPE at their worst site, and notes that the high cost of cloud tokens is pushing AI inferencing to the edge, where customers like the University of Texas at Dallas now run their network with one engineer instead of four.
Nokia

How AI Agents Drive Autonomous Networks

  • Nokia defines an AI agent as an autonomous element that takes a goal, perceives its environment, and continuously plans, decides, acts, and evaluates.
  • A March 2026 Appledore study finds operators expect AI to cut mean time to repair, optimize capacity, speed service activation, and reduce truck rolls.
  • With the industry centered on TM Forum autonomy Level 2, Nokia's agentic framework, including MCP and A2A, helps operators progress with less risk.
Abstract
Mike Thompson, VP, Head of Network Automation Product Management at Nokia, describes AI as an inflection point for service provider network operations and defines AI agents as autonomous elements that perceive their environment and continuously plan, decide, act, and evaluate to orchestrate across diverse networking domains with security and trust. Citing a March 2026 Appledore study and the TM Forum autonomous networks framework, which shows the industry currently centered around Level 2, Thompson explains how Nokia's agentic framework, including MCP and A2A, helps operators reduce mean time to repair, optimize capacity, boost productivity, and lower costs with less risk.
Ribbon, AWS

How Reasoning AI Is Reshaping Telecom Operations

  • Reasoning AI correlates thousands of alarms from 15 to 20 systems, turning hours or days of root-cause analysis into minutes.
  • Humans stay in control for the next couple of years through AI co-pilots for the NOC and network changes.
  • The milestone to watch is the first operator trusting closed-loop remediation to run unattended in production.
Abstract
Ram Ramanathan, VP Product Management – AI, Automation & Cloud at Ribbon, and Chris McKenna, Global Head of Business Development, Telecoms at Amazon Web Services, explain how reasoning AI helps operators manage rising 5G and AI complexity amid shrinking margins by correlating thousands of alarms across disparate systems to deliver root-cause answers in minutes rather than hours or days. They advocate a "crawl, walk, run" approach that keeps humans in control through NOC and network change co-pilots while the industry moves toward intent-based agentic automation, and they predict that within two to three years the first operator will trust closed-loop remediation to run unattended in production.
Download our 2026 AI for Networking report: Charting the Agentic Waters, and send us your feedback!
Download the 2026 AI for Networking Report

By downloading this report, you agree to our Terms of Use and Privacy Policy.

Continue exploring

More executive showcases across next-gen infrastructure.

2026

2026 MWC Barcelona

Leading executives at MWC 2026 in Barcelona discussing AI, ML, GenAI, agentic AI, AI-RAN, O-RAN, 5G-Advanced, 6G, and plenty more AI.

12 video interviews
2026

2026 OCP Global Summit

Exclusive videos from OCP Global Summit 2026 (Oct 12-15, San Jose). Topics: co-packaged optics, 800V racks, AI inference disaggregation, edge DPUs, AI computing continuum

28 video interviews
2026

2026 Technology Exploration Forum

Highlights from Ethernet Alliance's 2026 TEF. Topics: 224/448 Gbps/lane, 1.6/3.2 Tbps, AI fabric, InfiniBand, UCIe, CXL, PCIe Gen 6/7

20 video interviews
Browse all topics →