Fabrix.ai is introducing a multi-vendor, governed VibeOps operational intelligence platform designed to help IT, network, security and site reliability teams create operational dashboards, agents and automation workflows using natural-language instructions. The company is also adding domain-specific small language models, or SLMs, intended to improve the accuracy, performance, privacy and economics of operational AI.
The announcement reflects a broader transition. AI is accelerating application development, but the systems supporting those applications remain fragmented across observability, networking, security and IT service management. Fabrix.ai’s goal is to help operations teams keep pace while preserving production governance. Check out the full conversation and demo in the video below,
Extending Vibe Coding into Operations
Vibe coding has emerged as one of the most visible applications of generative AI. Developers can describe an intended outcome in natural language and use an AI coding assistant to generate or modify software. Fabrix.ai is extending that concept into operations, where the output may be a dashboard, an AI agent, an incident workflow or an automation rather than an application.
This distinction is important. VibeOps is not primarily aimed at turning network or IT operators into software developers. It is intended to help operational subject-matter experts translate their knowledge into useful tools without waiting weeks or months for a conventional development process.
“The beauty of vibe coding is essentially the ability to express an intent in natural language,” said Shailesh Manjrekar, chief marketing officer and head of AI strategy at Fabrix.ai. “The use cases remain the same. The outcomes remain the same. It is really how you achieve those outcomes that makes the difference.”
For enterprises, the potential benefit is greater operational agility. Network teams could create dashboards for VPN health or Wi-Fi performance, while SRE and security teams could develop agents for root-cause analysis or vulnerability exposure.
Governance Is the Critical Differentiator
Speed alone is not enough in enterprise operations. AI-generated code or automation can introduce security, reliability and cost risks if it is deployed without appropriate review. That is why Fabrix.ai places “governed” at the center of its positioning.
The platform provides a landing zone where AI-generated artifacts can be inspected, tested, version controlled and summarized before promotion into production, retaining human oversight and established change controls.
Governance also depends on grounding agents in accurate enterprise data. Most large organizations already operate numerous domain-specific platforms, and replacing or centralizing all of them is rarely practical. Fabrix.ai instead uses a data fabric to connect with existing APM, IT operations, network-performance, IT service-management and device-level data sources.
On top of those connections, the platform creates what Fabrix.ai calls a living ontology: a continuously updated semantic layer that identifies available data, its relationships and whether it is ready for use by an agent.
“We call it the GPS for AI agents,” Manjrekar explained. “The ontology layer is primarily targeted at providing accurate and curated information to the AI agent so it is not hallucinating.”
Agents need more than telemetry; they need context about applications, devices, dependencies, users and services. Fabrix.ai’s ontology is intended to provide that structure without moving enterprise data into another repository. Universal MCP-based tooling connects agents to APIs, platforms and devices, while an agent hierarchy directs tasks to specialists.
Small Models for Domain-Specific Work
Fabrix.ai is also introducing SLMs for platform interaction, AIOps and vulnerability exposure. Unlike a general-purpose cloud model, these smaller models can be tuned to a customer’s environment, service maps and correlation policies.
Keeping a model within the customer environment may improve data sovereignty and reduce network dependencies. Domain specialization can improve accuracy for defined tasks, while local execution may lower inference costs. Fabrix.ai plans to use an internal router to select an SLM or foundational model based on the task.
Enterprises should still validate these benefits in their own environments. Accuracy claims, operating costs and infrastructure requirements will vary by use case. The important point is that operational AI is moving toward a portfolio of models rather than assuming one large model should handle every task.
From Swivel-Chair Operations to Coordinated Action
Fabrix.ai demonstrated a network-health dashboard aggregating data from platforms including Cisco Catalyst Center, Meraki, NDFC, ThousandEyes, Splunk, ServiceNow and Juniper Mist. The dashboard presented device status, incidents, VPN health and affected sites through one operational view. Users could then modify or generate dashboards through supported coding assistants and review the resulting artifacts before deployment.
The company also described a Fortune 500 production deployment spanning approximately 25 tools, 12,000 to 13,000 assets and 207 applications. Fabrix.ai reported estimated annual savings of $1.23 million across root-cause analysis, VPN, Wi-Fi and related use cases. Those results are vendor-reported, but they illustrate the business case: reduce time spent collecting evidence from multiple consoles and redirect skilled personnel toward diagnosis, decisions and remediation.
The more meaningful shift is from reactive investigation to continuous operational assistance. Ambient agents can monitor conditions, correlate signals and identify issues before users open a ticket. Human operators can remain responsible for final approvals or remediation where risk requires it.
Why It Matters
Fabrix.ai’s announcement fits a broader industry movement from observability toward agentic operations. Dashboards help teams see problems; agentic systems are increasingly expected to investigate, recommend and eventually execute actions. However, enterprises will not trust these systems based on conversational interfaces alone.
Successful adoption will depend on trusted data, explainable reasoning, model and token-cost controls, role-based access, testing and human governance. It will also require a pragmatic deployment model. As Manjrekar advised, “Start with a smaller use case where you can identify tangible goals.” VPN health, Wi-Fi assurance or monitoring a critical application can provide a bounded environment for measuring productivity, accuracy and time to resolution.
Fabrix.ai is addressing a real operational challenge: application and AI environments are changing faster than traditional operations processes can comfortably support. Its opportunity is to demonstrate that governed VibeOps can consistently turn that speed into measurable enterprise outcomes without introducing another silo, another ungoverned automation layer or another expensive source of complexity.
See Fabrix.ai at Cisco’s Splunk .conf26
