Empirik Spins Out of Sequoia With $21M: The 'Infrastructure Compiler' That Predicts Outages Before They Happen

The most interesting infrastructure startup of the week wasn’t hatched in a garage. It was hatched inside Sequoia Capital’s own IT department — and on September 1, 2026, it spun out as an independent company with $21 million in seed funding from Sequoia, Canapi, and Alumni Ventures.

The company is called Empirik, and its mission sounds deceptively simple: predict and prevent tech outages before they happen, instead of alerting you after the damage is done.

The origin story: two operators inside Sequoia

Before joining Sequoia Capital in 2020 as its Chief Digital and Information Officer, Avon Puri spent more than a decade running infrastructure at scale — as SVP and CIO at Rubrik, and as VP of Engineering and Applications at VMware. Working alongside him at all three companies was Sudheer Dhurjati, another Sequoia IT leader.

Around three years ago, as large language models began showing their true potential, the two men drew a conclusion from their decades of shared experience: enterprise infrastructure is stuck in a reactive posture. Companies pour tens of billions of dollars into observability — metrics, logs, traces — but all of that tooling is inherently post-mortem. It tells you that latency spiked, that errors climbed, that a pod crashed. It rarely tells you why, and it almost never tells you before.

The insight that became Empirik’s founding pillar is that the real driver of infrastructure behavior is change: a new identity policy attached, a Terraform provider completing, a feature flag flipping in production. Yet change has historically been treated as administrative overhead — a ticket, an approval form, or tribal knowledge buried in Slack threads and email archives. It was recorded for compliance, but never put to work doing actual engineering.

Puri and Dhurjati built the first version of the product inside Sequoia. After putting it through the same evaluation any other Sequoia investment would face, the firm spun it out and installed a seasoned CEO: Kartik Chandrayana, a repeat founder who sold his first company, Twin Prime, to Salesforce, then spent five years as Salesforce’s VP of Product for Observability and Big Data, and most recently served as CPO at Quantum Metric.

What the product actually does

Empirik tracks system changes and infers their ripple effects across the entire infrastructure stack. Sequoia partner Bogomil Balkansky describes it as an autonomous “traffic cop” for the environment: it permits low-risk changes through automatically, sets guardrails on larger ones, and flags the most dangerous updates for human review before they ship.

Under the hood, Empirik calls its core engine an infrastructure compiler. Just as a traditional compiler translates source code into executable software, Empirik converts metadata from a customer’s clouds, identity providers, CI/CD pipelines, and SaaS systems into a live, unified operational graph — mapping thousands of relationships between assets and accounts. That graph is how the product knows, in real time, what changed, who changed it, and whether it’s the reason something broke.

Instead of telling you “latency in service X increased 300% in the last 24 hours,” Empirik tells you “Joe deployed a Terraform provider that shut down three VMs in AWS” — the actual cause and context, not just the symptom. And because risk is computed at the inception of a change, Empirik can warn or stop Joe even before he deploys.

The team deliberately avoided the common startup shortcut of building exclusively for greenfield Kubernetes or a single public cloud. From day one, Empirik was designed for the Fortune 500 reality: a multi-decade hybrid stack spanning public clouds, legacy on-prem infrastructure, identity layers, and critical SaaS applications. A graph that only sees half the environment is useless to an enterprise buyer.

Why now: agentic AI needs a map to reason over

The timing of the launch is not accidental. As Chandrayana frames it, infrastructure today is where software engineering was three years ago — before Claude Code and Cursor — when every action needed a human in the loop because no reliable, machine-readable map of the system existed for a model to reason over.

Empirik is building that map and putting it to work as what it calls the Autonomous Infra Engineer — AI that can safely plan, reason about, and operate enterprise infrastructure. The product already plugs into agentic workflows via MCP, connecting directly into tools like Claude and ChatGPT so that humans and AI agents alike can reason over infrastructure change with real context rather than guesswork.

The company’s stated ambition is to do for infrastructure engineers what Cursor and Claude Code did for software developers: automate the routine work so humans can focus on higher-value problems. As agentic AI accelerates the pace of software development itself, the volume of changes hitting production is exploding — making tools that can keep up with constant system change more vital than ever.

Traction: from TCBPay to Guardant Health

Although it only emerged from stealth this week, Empirik has been quietly shipping since earlier this year, and its customer list already ranges from startups to several Fortune 500 players — including S&P Global, Guardant Health, and a major consumer packaged goods company.

The case studies Sequoia published alongside the announcement are concrete:

• TCBPay, a payment processor handling more than $1.5 billion in annual volume, now catches sensitive configuration-file changes in 10 seconds, down from 30 minutes.
• Avahi, a managed service provider, embedded Empirik into its GitHub Actions to evaluate and block risky changes at the source.
• Guardant Health found the product valuable enough to expand from production to all development environments — and even to on-premises.
• A Fortune-50 CPG company is leaning on Empirik to drive most of its agentic and autonomous infrastructure roadmap, after a P0 outage caused by an unapproved change first drew it to the product.

Empirik says it now processes millions of raw change and telemetry events every week, over dependency graphs spanning a few million resources across all major clouds and on-prem systems. Customers report benefits across preventing risky changes from reaching production, detecting unapproved changes, faster incident response, rollback decisions, and drift detection.

A new category — or a missing pillar?

Balkansky argues that Empirik is in a category of its own for now, positioning it as a complementary layer to AI SRE platforms like Resolve and Sequoia-backed Traversal rather than a competitor. The framing is bold: observability is already a massive, proven market — Datadog, Dynatrace, and Splunk generate billions in revenue — but all of it is built to measure symptoms. Empirik’s bet is that there is a fourth pillar hiding in plain sight: the changes themselves.

That framing also explains the company’s larger ambition. Most enterprises today cannot answer a simple but critical question: what changed, and did that cause the incident? Empirik wants to be the system of record for that answer — for the humans asking it today, and for the AI agents that will increasingly be making those changes themselves tomorrow.

Whether “change intelligence” becomes a durable category or gets absorbed into the observability giants’ platforms remains to be seen. But with Sequoia’s incubation pedigree, an operator-heavy founding team, Fortune 500 logos already in production, and a $21 million war chest, Empirik has earned the right to make its case — and the outage you never had may be its best advertisement.