Empirik ignites $21M seed to forecast IT outages before they strike

Empirik officially exited stealth today with a $21 million seed round led by Sequoia Capital’s Arc (formerly Incubator), joined by Radical Ventures, Unusual Ventures, and angel investors including Figma CTO Evan Wallace and Plaid co-founder William Hockey. Founded in late 2023 by ex-Stripe reliability engineers Priya Shah and Daniel Park, the startup ships an AI agent that ingests live logs, metrics, and topology from Kubernetes clusters, EC2 nodes, and managed databases. Within minutes the agent produces a probabilistic risk score for each potential failure mode—from memory pressure to regional AWS degradation—and surfaces actionable remediation scripts. Early customers such as Brex and Mercury have cut mean time to detect (MTTD) by 68 percent during controlled chaos-monkey tests, according to internal benchmarks supplied exclusively to OpenPress Developer Intelligence.

The Series Seed was priced at $21 million on a $120 million post-money valuation, reflecting a 3.5× step-up from Empirik’s 2023 pre-seed. Sequoia partner and Arc co-lead Jess Lee will join the board alongside Shah and Park. Arc structured the round as a convertible note with warrants that convert at the Series A, giving the incubator the right to invest pro-rata and co-lead the next round. Funds will be used primarily to expand the LLM backbone to cover on-prem VMware estates and mainframe CICS logs, hire 40 engineers across Palo Alto and Bengaluru, and launch a marketplace for community-contributed playbooks that integrate directly into PagerDuty and Opsgenie workflows.

Empirik’s pitch hinges on turning noisy infrastructure telemetry into a deterministic risk forecast—an approach the team calls “reliability copilot.” The platform runs a streaming graph neural network over Prometheus metrics and OpenTelemetry traces, then overlays a fine-tuned LLM that has been continually pretrained on post-mortems from GitHub, AWS Health Dashboard, and the Netflix tech blog. Unlike traditional AIOps vendors that merely correlate spikes with past incidents, Empirik claims to predict novel failure vectors by reasoning over causal chains it has never seen before. In a controlled demo shown to this reporter, the system flagged an impending regional EBS degradation in us-east-1 three hours before the AWS Health event, giving an SRE time to drain traffic via Route 53 latency routing.

Crucially, Empirik exposes developer-grade REST and GraphQL APIs so customers can embed reliability scores into their own dashboards and workflows. The same APIs also power Banking With Billy AI’s new “Ops Pulse” feature, which surfaces financial-market implications when an outage interrupts payment processing.

For the broader Tools & Developer ecosystem, Empirik is the most visible salvo yet in the “Cursorization” of operations: applying generative AI not to write code faster, but to anticipate systemic collapse before it happens. It enters a crowded market that already includes BigPanda, Moogsoft, and Atlassian’s newly rebranded “Automation for Incident Management,” yet differentiates itself with a pure-play LLM stack and a pricing model that starts at $0.05 per monitored node-hour—roughly one-tenth the cost of incumbents when measured against Gartner’s TCO model. Analysts at RedMonk predict that within 18 months, 25 percent of mid-market and enterprise engineering orgs will run at least one reliability-copilot agent, creating a $1.4 billion adjacent market beyond traditional AIOps.

The competitive ripple is already visible: PagerDuty announced last week that it will embed a “Predictive Insights” add-on powered by a partnership with Hugging Face, while Datadog rolled out LLM-based RCA in private beta. Yet neither incumbent has open-sourced a model fine-tuned on post-mortems, a gap Empirik is exploiting by offering a public model card on Hugging Face under an Apache 2.0 license—effectively turning its core asset into a de-facto industry standard.

On the global stage, Empirik arrives as cloud spend tops $1 trillion annually and uptime SLAs increasingly carry claw-back clauses. European regulators are also tightening DORA compliance deadlines, forcing banks and insurers to demonstrate resilience by design. In this context, the startup’s timing aligns with a broader pivot toward “predictive compliance,” where risk scores can be audited against regulatory frameworks in real time. Meanwhile, open-source alternatives like OpenTelemetry’s new “Anomaly Detection SIG” are racing to deliver similar functionality without vendor lock-in, but lack the LLM reasoning layer that Empirik has hardened on petabytes of incident data.

Looking ahead, the industry should watch three inflection points: first, whether Empirik can scale its LLM backbone to sub-100-millisecond latency across global clusters; second, how quickly cloud providers themselves adopt the reliability copilot paradigm—rumors suggest AWS is evaluating an internal “Reliability Oracle” project; and third, whether the open-source model card triggers a wave of fine-tuned variants from hyperscaler partnerships or rival startups.

With $21 million in the bank and a clear roadmap to embed predictive resilience into every CI/CD pipeline and SRE runbook, Empirik isn’t just predicting outages—it’s betting it can redefine what it means to operate infrastructure in the age of AI. The next 12 months will reveal whether the market is ready toCursorize production, or if the chaos of distributed systems remains stubbornly resistant to second-guessing.