Empirik is spinning out as an independent company with $21 million in seed funding for software designed to predict infrastructure outages before system changes trigger them.
Sequoia Capital incubated the startup in 2023 and invested in the new round alongside Canapi and Alumni Ventures. Empirik tracks changes across a company’s technology environment, infers how they may affect dependent systems and decides which updates can proceed automatically or require human review.
The company is entering an observability market usually focused on detecting and resolving incidents. Its central claim is that AI can move infrastructure teams earlier in the cycle, from reacting to failures toward estimating the risk of a change before deployment.
An infrastructure map built around change
The idea originated with Avon Puri and Sudheer Dhurjati, technology leaders at Sequoia. Puri had spent more than a decade running infrastructure at Rubrik and VMware before joining the investment firm in 2020 as chief digital and information officer.
Three years ago, as large language models demonstrated broader capabilities, Puri and Dhurjati concluded that AI could help engineers prevent outages. They developed a system that observes changes and models the effects those changes could have across an interconnected estate.
Sequoia recruited Kartik Chandrayana as CEO earlier this year. Chandrayana previously served as chief product officer at Quantum Metric and led observability work at Salesforce.
Sequoia partner Bogomil Balkansky described Empirik as an autonomous traffic controller. Low-risk changes can pass, larger ones receive guardrails, and the most dangerous are escalated for a person to assess. That description comes from an investor, and TechCrunch’s report does not include independent measurements of prediction accuracy or prevented downtime.
The intended users are DevOps and site-reliability engineering teams. Empirik says automating routine troubleshooting allows those specialists to concentrate on higher-value work rather than investigating every change manually.
Early customers test the prevention thesis
Empirik launched earlier this year and says it has already signed customers ranging from startups to Fortune 500 businesses. Named users include S&P Global and Guardant Health, along with a large consumer packaged-goods company that was not identified.
The customer list shows early commercial adoption but not yet the outcomes of deployment. The source does not report how many incidents the platform predicted, the false-positive rate or whether customers reduced downtime after installation.
Those measures will be central because infrastructure is full of dependencies that may not be visible from a single monitoring source. A tool that blocks too many safe changes can slow engineering teams, while one that misses a critical relationship may fail to prevent the outage it was meant to anticipate.
Empirik says it complements AI site-reliability platforms such as Resolve and Sequoia-backed Traversal. Balkansky argues that the startup currently occupies its own category, although that is an investor’s assessment rather than an established market definition.
Faster software raises the operational stakes
Chandrayana compares Empirik’s ambition with the effect of Cursor and Claude Code on software development: automating selected work so specialists can move faster. In his framing, the company wants to bring agentic AI to infrastructure engineering in the way coding tools brought it to developers.
The connection is important to Empirik’s case. If AI-assisted development increases the speed and volume of software changes, operations teams must evaluate more potential effects across production systems. Faster creation can become a reliability problem when deployment controls and dependency knowledge do not scale with it.
The $21 million round gives Empirik resources to turn an incubated product into an independent business. It also places the company’s prevention claims under a clearer commercial test.
Existing observability tools already absorb large budgets for keeping services online, according to Balkansky, but he says most do not understand complex dependencies. Empirik is betting that a change-centered view can fill that gap.
Success will depend less on describing an autonomous infrastructure engineer than on showing reliable judgment: which updates are safe, which need constraints and which should stop. In systems where a small change can cascade across services, accurately predicting that boundary is the product.
