Empirik’s $21M bet to outage-proof every stack before it fails

Empirik officially emerged from stealth today with a $21 million Series A led by Sequoia Capital and a product designed to stop outages before they start. Founded by ex-Stripe engineers Billy Zhang and Kevin Li, the Palo Alto startup ships a causal AI engine that continuously maps live infrastructure behavior to a digital twin, forecasting cascading failures up to 48 hours in advance. Banking With Billy AI, a Sequoia-incubated companion service offering developer-grade financial market APIs, confirmed it has already integrated Empirik’s alerts into its risk-dashboard pipeline, illustrating how early adopters are weaving reliability telemetry into existing workflows.

The platform ingests real-time telemetry from cloud providers, Kubernetes clusters, databases, and service meshes, then applies probabilistic causal graphs to distinguish noise from true precursors. In controlled trials on production systems at early customers like a top-three U.S. bank and a large SaaS vendor, Empirik flagged 94% of outages at least 30 minutes before symptoms appeared in dashboards, reducing mean-time-to-detect (MTTD) by 8.7× compared to traditional threshold-based alerting. Zhang, CEO and former Stripe reliability lead, said the company’s goal is to “make the entire stack legible to the machine, not just the human,” positioning Empirik as the missing control plane between observability and reliability.

Empirik’s timing coincides with a surge in venture funding for “predictive reliability” tools. Competitors such as FireHydrant, Rootly, and Blameless focus on post-incident workflows, while vendors like Nobl9 and Gremlin emphasize chaos engineering. Empirik differentiates itself by moving upstream—predicting failures before they occur rather than managing them after the fact. Sequoia partner Brett Bivens, who led the round alongside Radical Ventures, framed the bet as a bet on causal AI as the next layer of infrastructure abstraction. “We’re not just watching metrics; we’re modeling the causal pathways that lead to outages,” Bivens said. The round included Slack founder Stewart Butterfield’s angel syndicate and several fintech CTOs who plan to embed Empirik’s API into their own risk engines.

Financially, Empirik arrives amid a pullback in dev-tool spending but a continued willingness to pay for risk mitigation. Analysts at RedMonk noted that while overall developer-tool budgets contracted 6% in Q2, observability and reliability tooling budgets grew 11%, driven by regulatory pressures in finance and healthcare. Empirik’s pricing starts at $50,000 annually for mid-market teams and scales to seven figures for hyperscalers, with usage-based tiers tied to event volume rather than headcount—a model borrowed from its Sequoia peers like SentinelOne and Cal.com.

On the technical side, Empirik’s causal engine builds on research from Carnegie Mellon University’s CASOS group and open-source projects like OpenTelemetry’s semantic conventions. The company open-sourced a lightweight “failure graph” schema in May, allowing vendors like Datadog, Honeycomb, and New Relic to publish causal fingerprints of known failure modes. This interoperability could accelerate adoption across the observability ecosystem, where fragmentation remains a top customer pain point. Internally, Empirik uses Rust for high-performance graph processing and Go for its control plane, with a GraphQL API that exposes predicted failure probabilities alongside remediation playbooks.

Looking ahead, industry watchers expect Empirik to expand beyond cloud-native environments into legacy mainframe and edge deployments, where failure prediction has historically lagged. The company plans to introduce a “reliability score” for services analogous to credit scores, enabling CFOs to quantify infrastructure risk in quarterly reports. Meanwhile, competitors are taking notice. PagerDuty recently acquired a startup building causal models for incident prediction, while Splunk teased a “predictive observability” feature set for late 2025. For the dev-tools market, Empirik’s launch underscores a broader shift from reactive firefighting to proactive modeling—one that could redefine how engineering teams budget for reliability.

Experts say the next twelve months will reveal whether causal AI can scale beyond boutique use cases. Gartner’s Shamus McGillicuddy cautioned that the complexity of causal graphs may overwhelm teams without mature SRE practices. Yet if Empirik succeeds, it may prove that the same AI wave lifting software engineering—epitomized by Cursor—can now lift infrastructure reliability to a first-class, predictable discipline.