Modern enterprise infrastructure has long operated in a reactive posture. For decades, companies have poured tens of billions of dollars into observability tools—spanning metrics, logs, and traces—to tell them when a complex system is failing. Yet, traditional observability is inherently post-mortem, designed to measure symptoms after the damage has already been done.
Today, a new startup called Empirik is emerging from stealth with $21M in funding to tackle this foundational flaw, introducing what it calls the missing fourth pillar of enterprise observability: tracking the changes themselves.
The company was incubated internally at Sequoia Capital, born from an observation shared by industry veterans who watched the same operational cycle repeat across multiple large tech organizations. Avon Puri, Sequoia’s Chief Digital and Information Officer, spent two decades running infrastructure at scale as SVP and CIO at Rubrik and VP of Engineering and Applications at VMware. Alongside Sudheer Dhurjati, who worked with Puri across three major companies, the pair observed that enterprise IT departments were continually caught playing firefighting games.
They realized that the true driver of infrastructure behavior is change itself. Whether an engineer attaches a new identity policy, completes a new Terraform provider configuration, or flips a feature flag in a live production environment, change is what dictates how a system performs. Historically, however, change has been treated merely as an administrative process constraint—a static ticket, an approval form, or tribal knowledge buried in email threads and Slack channels. While sometimes recorded for compliance purposes, change data has rarely been put to work to execute actual infrastructure engineering jobs.
Empirik’s founders recognized that by capturing and understanding the real-time stream of changes across every layer of an environment, organizations can move beyond better alerts and unlock true autonomy. Every change reflects an engineer’s intent. Once a system can interpret that intent and accurately assess its impact, it can autonomously decide whether to act or loop in a human operator. By bringing agentic capabilities to infrastructure, Empirik aims to solve the full lifecycle of infrastructure engineering challenges, shifting organizations from a reactive firefighting stance to a proactive operational posture.
Why Change, and Why Now
The timing of Empirik’s market debut coincides with a massive shift toward agentic automation in software development. Agentic coding has taken off rapidly in recent years because code lives in version-controlled repositories, making a system’s underlying architecture and intent entirely legible to machine learning models.
Infrastructure, however, has never possessed that foundational advantage. While Infrastructure-as-Code (IaC) represents a step forward, it typically expresses a declared state rather than capturing what is actually running dynamically in production. Consequently, it remains inherently out of step with reality. There has historically been no unified, real-time representation of an organization’s infrastructure spread across the maze of cloud resources, Kubernetes clusters, networking protocols, identity layers, and legacy systems. Unlike application code, infrastructure has largely lacked the clean notion of a pull request. Every change has continued to rely heavily on institutional memory and the tribal knowledge of human operators rather than the rigorous automation of software.
While established observability giants like Datadog, Dynatrace, and Splunk generate billions in revenue by measuring symptoms—such as a sudden spike in latency, climbing error rates, or a crashing pod—they do not capture the root cause. Empirik’s core thesis is that the fourth pillar hiding in plain sight is the actual change event. Instead of alerting a team that latency in a specific service increased by 300 percent over a 24-hour period, Empirik pinpoints the exact cause and context, informing engineers that a specific team member deployed a Terraform provider that inadvertently shut down virtual machines. Moreover, the platform is designed to warn or even block risky deployments before they ever reach production by computing risk at the very inception of the change.
Building the Infrastructure Compiler
To realize this vision, Empirik is led by CEO Kartik Chandrayana, a repeat founder whose previous company, Twin Prime, was acquired by Salesforce. Chandrayana subsequently spent five years as Salesforce’s VP of Product for Observability and Big Data, followed by a stint as Chief Product Officer at Quantum Metric.
According to Chandrayana, enterprise infrastructure today resembles software engineering from three years ago—before the advent of advanced coding assistants, when every administrative action required human intervention because no reliable, comprehensive map of the system existed to reason over. Empirik is building that missing map to function as an Autonomous Infra Engineer, providing AI that can safely plan, reason about, and operate enterprise infrastructure. The platform is already integrated into agentic workflows via the Model Context Protocol (MCP), connecting directly into tools like Claude and ChatGPT so that both humans and AI agents can evaluate infrastructure changes using real context rather than guesswork.
Under the hood, the company relies on what it terms an infrastructure compiler. Just as a traditional software compiler translates source code into executable software, Empirik’s engine converts raw metadata harvested from a customer’s multi-cloud environments, identity providers, CI/CD pipelines, and SaaS systems into a live, unified operational graph. This graph maps thousands of complex relationships spanning assets and user accounts, allowing the product to determine in real time precisely what changed, who initiated the change, and whether it triggered a system failure.
Crucially, the founding team designed Empirik to accommodate the messy realities of Fortune 500 enterprises. Rather than building exclusively for greenfield Kubernetes environments or a single public cloud vendor, the platform handles complex, multi-decade hybrid enterprise stacks encompassing public clouds, legacy on-premise infrastructure, identity management layers, and critical third-party SaaS applications. By connecting the entire surface area of an organization, Empirik provides teams and AI agents with complete operational context.
Proving the Thesis in Production
The market validation for Empirik’s approach is already visible across major enterprise deployments. Interest from Fortune 500 financial institutions often stems directly from the desire to prevent severe outages caused by unapproved or poorly understood configuration changes.
For instance, payment processing firm TCBPay, which handles over $1.5 billion in annual volume, now catches sensitive configuration file changes in just 10 seconds, a dramatic improvement from the previous duration of 30 minutes. Managed service provider Avahi integrated Empirik directly into its GitHub Actions workflows to evaluate and block risky operational changes at the source. Meanwhile, Guardant Health expanded its deployment from production systems to cover all development environments as well as on-premise infrastructure, and a global Fortune 50 consumer packaged goods enterprise is expanding its footprint to drive its autonomous infrastructure roadmap.
Today, Empirik processes millions of raw change and telemetry events every week, tracking dependency graphs that span millions of resources across major cloud providers and legacy on-premise systems. Customers are reporting tangible benefits across multiple domains, including preventing risky changes from reaching production, detecting unauthorized modifications, accelerating incident response times, improving rollback decisions, and catching configuration drift.
As enterprises continue to struggle with the deceptively simple question of what changed and whether it caused an incident, Empirik is positioning itself as the foundational system of record—designed for human operators today, and built for the autonomous agents that will increasingly manage enterprise systems tomorrow.
Comments (0)