Forward Networks Unveils Advanced Agentic AI for Network Ops

Revolutionizing Network Operations with Agentic AI

Network operations teams are increasingly encountering a plethora of AI-driven tools designed to streamline management through natural language interfaces. While many of these tools primarily serve as conversational wrappers around existing functionalities, offering convenience, they often fall short of fundamentally transforming how networks are managed and verified. Forward Networks initially entered this space in 2024 with its AI Assist feature, a conversational interface for network queries.

However, in 2026, Forward Networks is taking a distinct approach by introducing an agentic AI system that transcends mere conversational interaction. This new system moves beyond simply answering questions; it dynamically formulates and executes multi-step workflows across diverse hybrid and multi-cloud landscapes. Crucially, it maintains mathematical verification of its recommendations, ensuring accuracy and reliability.

Forward Networks, established in 2013 by four Stanford PhDs, has dedicated over a decade to developing mathematically precise network digital twin technology. The integration of these new agentic capabilities into Forward AI is deeply rooted in the company’s foundational vision, which centers on delivering an accurate and actionable representation of the entire network infrastructure.

David Erickson, CEO and co-founder of Forward Networks, emphasized the strategic question that guided the company’s inception. “When we founded Forward Networks, we started with a simple but strategic question: Does the network actually behave as intended, by design, in production, and across changes?” Erickson noted to Network World. He further explained that answering this profound question demands more than mere visibility; it necessitates a mathematically accurate model of the network itself, a core component of their digital twin technology.

From AI Assist to Agentic Intelligence

Forward AI represents a significant leap from the company’s earlier AI Assist feature, which primarily translated English questions into queries against a normalized data model. The latest system functions as an intelligent agent capable of dynamically constructing multi-step execution plans. These plans can interact with both the internal digital twin and external systems, such as ServiceNow, demonstrating a sophisticated level of automation and integration.

Nikhil Handigol, co-founder and chief AI officer at Forward Networks, explained the system’s enhanced capabilities to Network World. “It’s a conversational agentic system that is designed to simplify operations overall,” Handigol stated, highlighting its comprehensive approach to streamlining network management. This advanced system can interpret complex requests and orchestrate a series of actions to achieve a desired outcome, moving beyond simple question-and-answer interactions.

The practical advantages of this agentic approach become particularly evident in troubleshooting scenarios. For instance, when tasked with triaging a ServiceNow ticket, the agent performs a comprehensive sequence of actions. It begins by analyzing the ticket content, then gathers relevant contextual information about mentioned entities from the digital twin. Subsequently, it automatically performs path traces to diagnose connectivity issues and ultimately returns a detailed diagnosis. Throughout this entire workflow, the complete process remains transparent and visible to human operators, allowing for oversight and validation.

This method stands in stark contrast to basic natural language query systems, where a question is translated, and an immediate answer is returned. The agent’s ability to build and execute a sophisticated plan involving multiple data sources and analytical steps marks a substantial evolution. It demonstrates an understanding of operational tasks that extends far beyond simple information retrieval, offering a more proactive and integrated solution for network management challenges.

Custom Framework for Enhanced Context Control

Forward Networks made a strategic decision to develop its own agentic framework rather than relying on existing tools like LangChain or CrewAI. This choice was driven by the imperative to maintain precise control over context engineering, a factor Handigol identified as the primary engineering challenge for agentic AI systems. Ensuring that the AI agent receives exactly the right amount of relevant information, without extraneous data, is critical for accurate and efficient operation.

“We built our own because we wanted complete control over how the agent executes,” Handigol explained. He further elaborated that the core engineering problem revolves around context engineering—how to define and maintain the necessary context for the agent to effectively interact with the Large Language Model (LLM) and derive correct answers. This bespoke framework allows for fine-grained management of information flow, which is crucial for preventing errors and improving the quality of AI-generated responses.

The team defines effective context engineering as providing all pertinent information while rigorously eliminating excess noise. Supplying insufficient information can lead to incorrect answers, while an overload of irrelevant data can distract the model from its intended task. Striking this delicate balance is paramount for the agent’s performance and reliability.

Contextual information is meticulously drawn from Forward Networks’ hierarchical data stack, a robust foundation that supports the agentic system. At the base layer, raw configuration data, current state information, and operational statistics are directly collected from network devices. The subsequent tier normalizes this raw data into a comprehensive, queryable model that provides a detailed view of everything present in the network and its current configuration. This structured approach ensures that the agent has access to accurate and up-to-date network intelligence.

The top tier of this data stack performs sophisticated behavioral analysis. This layer is capable of answering complex questions, such as tracing arbitrary packet paths through the network, assessing asset exposure to external networks, and calculating the lateral movement blast radius if certain assets become compromised. Furthermore, the system intelligently correlates network data with information from Configuration Management Database (CMDB) systems, IP Address Management (IPAM) tools, and various ticketing systems. This holistic approach to context ensures the agent can make informed decisions based on a wide array of interconnected data sources.

Mitigating Hallucinations and Ensuring Transparency

The inherent probabilistic nature of large language models, which predict the next token based on learned patterns, introduces a fundamental risk of “hallucinations.” In the context of network operations, where a single misconfiguration can precipitate widespread outages, such risks are unacceptable. Forward AI employs a dual-pronged strategy to effectively mitigate this critical problem, prioritizing both accuracy and operational integrity.

First, the system’s recommendations are firmly anchored in a deterministic digital twin, rather than solely relying on probabilistic predictions generated by the LLM. This digital twin provides a mathematically accurate representation of the network, ensuring that any actions or diagnoses proposed by the AI are grounded in verifiable facts. This foundational reliance on a deterministic model significantly reduces the potential for erroneous outputs that could arise from purely probabilistic reasoning.

Second, Forward AI is designed for complete transparency in its reasoning processes. All evidence used to substantiate the agent’s recommendations is directly accessible to operators during their interactions with the system. This crucial feature empowers network personnel to validate the agent’s logic and underlying data before implementing any proposed actions. Such transparency fosters trust and provides a critical layer of human oversight, transforming the AI into a collaborative tool rather than an opaque decision-maker.

To further ensure the reliability and continuous improvement of the agentic system, Forward Networks employs a rigorous evaluation framework throughout its development cycle. As the agent’s code evolves and new capabilities are introduced, this framework is meticulously designed to prevent any regression in existing functionalities. Concurrently, it systematically validates improvements in new areas, ensuring that enhancements do not inadvertently compromise established performance. This continuous evaluation process is fundamental to maintaining the high standards of accuracy and dependability required for enterprise-grade network operations.

Multi-Vendor Support Across Hybrid Environments

The digital twin, which forms the bedrock of Forward AI, boasts extensive support for dozens of network vendors and encompasses a wide range of protocols, from Layer 2 through Layer 7. This comprehensive compatibility extends across diverse infrastructure environments, including on-premises setups, major cloud providers such as AWS, Azure, and Google Cloud, and even modern Kubernetes deployments. This broad and deep integration capability is a significant differentiator, allowing the agentic system to operate seamlessly across complex, heterogeneous networks.

“The major advantage that Forward Networks has is that we are multi-vendor,” Erickson reiterated, emphasizing the breadth of their platform. He added, “We support all network vendors that are out there, and we make the lives better for the customers of all of these vendors.” This vendor-agnostic approach is critical in today’s multi-vendor enterprise landscapes, where organizations often utilize equipment and services from various providers.

This extensive multi-vendor support eliminates a significant source of operational inefficiency and “toil” for network operators. Instead of needing to log into numerous vendor-specific command-line interfaces (CLIs), learning distinct command syntaxes for each, and manually normalizing data from disparate sources, operators can leverage a unified platform. Forward AI streamlines these tasks by providing a single pane of glass for managing and verifying network behavior across the entire infrastructure.

The comprehensive breadth of protocol and technology support is absolutely essential for the agentic system to function reliably and effectively. Without thorough data collection and analysis from every corner of the network, any recommendations or actions proposed by the AI would be plagued by gaps, potentially leading to critical errors or incomplete solutions. Such comprehensive coverage ensures that the agent always operates with a complete and accurate picture of the network’s current state and intended behavior.

Handigol underscored the importance of this robust foundation. “Everyone is building some sort of AI-powered capability for their own systems,” he observed, acknowledging the widespread adoption of AI. He then highlighted the key differentiator: “What differentiates one AI-powered capability from the other is the foundation on top of which it is built.” This underscores Forward Networks’ commitment to building its AI solutions on a strong, mathematically verified digital twin, ensuring unparalleled accuracy and reliability in network management.