Empirik
by Empirik (empirik.ai)
Change observability for infrastructure — compute the blast radius before the change lands
Empirik is an AI change observability platform for platform, DevOps and SRE teams whose infrastructure changes now move faster than human review can keep up with. It continuously models the live environment across cloud, Kubernetes, IAM and CI/CD, computes the blast radius of a proposed change before it ships, then auto-approves it, puts guardrails on it, or escalates it to a human.
Empirik is a change observability platform that sits in front of infrastructure changes rather than behind incidents. The company emerged from stealth on 2 September 2026 with $21 million from Sequoia Capital, S32, Canapi Ventures and Alumni Ventures, after being incubated inside Sequoia from 2023 by Avon Puri — formerly Sequoia's chief digital and information officer, and before that an infrastructure leader at Rubrik and VMware — and Sudheer Dhurjati. Kartik Chandrayana, previously chief product officer at Quantum Metric and head of Salesforce's observability business, joined as chief executive in early 2026. The product continuously models a customer's live application environment across public cloud, on-premises systems, Kubernetes, virtual machines, IAM policies, CI/CD pipelines and SaaS applications, treating access control lists, routing tables and permissions as first-class infrastructure objects rather than configuration detail. When an engineer opens a pull request, change ticket or pipeline run, Empirik captures the intent behind it, projects that mutation against the live model, and computes the exact blast radius before the change lands. It then acts as an autonomous gate: low-risk changes are approved automatically, larger changes get guardrails, and the riskiest are escalated for human review. The platform also reconciles intended, deployed and runtime state to trace configuration drift to its source and identify orphaned or idle infrastructure and its owner. Named production customers include Guardant Health, Avahi Systems and TCBPay, alongside a Fortune 50 consumer packaged goods company and S&P Global.
The VP of Platform Engineering or Head of SRE at an enterprise where AI coding agents have raised change volume faster than the change-approval process can absorb it, and where change-induced outages — not unknown bugs — are the main source of downtime.
Every infrastructure change arrives at review with its computed blast radius attached, so low-risk changes stop queueing behind human approval and genuinely risky ones stop slipping through unexamined.
At a Glance
- Category
- Infrastructure & Cloud
- Pricing
- Contact for pricing
- Target Market
- CTOs, CIOs, VP Engineering, Platform Engineers, Site Reliability Engineers, DevOps Teams
- Deployment
- Cloud-first, Multi-cloud
- Founded
- 2023
- Headquarters
- San Francisco, United States
Key Features
- ✓Intent capture at the change source
Reads the engineer's intent directly from the pull request, change ticket or pipeline that originates the change, rather than inferring it from the diff after the fact.
- ✓Blast radius computation
Projects a proposed mutation across the live environment model and computes which dependent services it would touch, before the change is applied to production.
- ✓Autonomous change gating
Approves low-risk changes automatically, attaches guardrails to larger ones and escalates the riskiest for human review, so approval throughput scales with change volume.
- ✓Continuous environment model
Maintains a live graph spanning public cloud, on-premises systems, Kubernetes, virtual machines, IAM, CI/CD and SaaS, so dependencies stay current between deploys.
- ✓IAM and network objects as first-class citizens
Models IAM policies, access control lists and routing tables as infrastructure objects, catching permission and routing changes that config-file scanners miss entirely.
- ✓State reconciliation and drift tracing
Compares intended, deployed and runtime state across the stack and traces every divergence back to its source change and operational impact.
- ✓Orphaned infrastructure discovery
Finds idle and unowned resources, identifies who owns them, and establishes a safe removal path instead of leaving them to accrue cost and risk.
Capabilities
Use Cases
- •Pre-merge infrastructure review
A platform engineer opens a Terraform pull request; Empirik projects it against the live environment and reports which downstream services the change would affect before merge.
- •Safe autonomy for AI coding agents
An AI agent proposes an infrastructure change; Empirik supplies the system context and risk verdict that decides whether it can execute without a human in the loop.
- •IAM permission change control
A routine policy edit is evaluated against the access graph so an over-broad grant or a severed service-to-service permission is caught before it reaches production.
- •Configuration drift investigation
When runtime state stops matching what was deployed, the platform reconciles the three states and traces the divergence back to the specific change responsible.
- •Idle infrastructure cleanup
Finance or platform teams identify orphaned resources, establish who owns each one, and decommission them through a path that will not break a dependent service.
Ideal For
Best For
- ✓Gating Terraform, Kubernetes and IAM changes on a computed impact analysis instead of a checklist-based CAB review
- ✓Preventing change-induced outages in multi-cloud estates where dependency graphs are too large to hold in an engineer's head
- ✓Giving AI coding agents enough system context that the infrastructure changes they propose can be evaluated automatically
- ✓Reconciling intended, deployed and runtime state to find configuration drift and trace it back to the change that caused it
- ✓Identifying orphaned and idle infrastructure, establishing ownership, and building a safe decommissioning path
Not Ideal For
- ✗Teams looking for incident response or root-cause analysis after an outage has started — Empirik's own founders position it as complementary to AI SRE tools such as Resolve AI and Traversal, not a replacement
- ✗Small engineering teams on a single cloud with a handful of services, where the dependency graph is simple enough that a human reviewer already sees the blast radius
- ✗Buyers who need published pricing, a self-serve trial or a documented compliance posture before starting an evaluation — none of these are public as of September 2026
- ✗Organisations that cannot grant a third-party platform broad read access to IAM, CI/CD and production configuration, which the change model requires to work
Deployment
Market Analysis
Pros
- ✓Unusually strong founding and operating bench for a seed-stage company: a former Sequoia CDIO with Rubrik and VMware infrastructure experience, plus a CEO who ran Salesforce's observability business and was CPO at Quantum Metric
- ✓Named production references at launch across regulated and demanding buyers — Guardant Health, S&P Global, a Fortune 50 CPG company, Avahi Systems and TCBPay — rather than an unnamed 'design partners' claim
- ✓Addresses a genuinely underserved position in the stack: existing observability answers 'what broke', while change risk is still handled by human review boards
- ✓Coverage spans on-premises and multiple public clouds plus Kubernetes and SaaS, so it is not limited to a single hyperscaler's estate
Cons
- ✗No independent user reviews exist anywhere — no G2, Capterra, TrustRadius or PeerSpot listing, and a Hacker News search returns no discussion of the company, so every capability claim currently traces back to the vendor or its launch press
- ✗No published pricing, no free tier and no self-serve trial, so evaluation cannot start without a sales cycle and cost cannot be modelled in advance
- ✗The company only became independent on 2 September 2026 after three years of incubation inside its lead investor; deployment breadth, scale limits and false-positive rates are unproven in public
- ✗Requires broad read access across cloud accounts, IAM, CI/CD and SaaS to build its model, which is a substantial security-review burden and a hard sell in tightly segmented environments
- ✗No compliance certifications (SOC 2, ISO 27001) are published, which will stall procurement at regulated buyers despite the healthcare and financial-data logos already cited
Pricing
Enterprise
Contact for pricing
- ✓Continuous environment modelling across cloud, on-prem, Kubernetes, IAM, CI/CD and SaaS
- ✓Blast radius computation on pull requests, change tickets and pipelines
- ✓Governed autonomous execution with human escalation for high-risk changes
- ✓Drift reconciliation and orphaned infrastructure discovery
No list pricing is published anywhere, and there is no self-serve tier or public trial — the company launched from stealth on 2 September 2026 selling directly to platform and SRE organisations at Fortune 500 accounts, so every deal goes through sales. Nothing is disclosed about whether the meter is per environment, per change, per node or per seat, which makes it impossible to model cost before a vendor conversation. Budget for a procurement and security-review cycle as well, since the platform requires broad read access across cloud accounts, IAM and CI/CD.
Security & Compliance
Sources
This page was written from 6 sources, 6 on domains other than empirik.ai.
- 1.techcrunch.com — sequoia incubated empirik launches with 21m to predict outag
- 2.prnewswire.com — empirikai emerges from stealth with 21 million to build the
- 3.pulse2.com — empirik ai raises 21 million seed round to build autonomous
- 4.tekedia.com — sequoia backed empirik raises 21m to use ai to prevent infra
- 5.aijourn.com — empirik ai emerges from stealth with 21 million to build the
- 6.thesaasnews.com — empirik raises 21m seed
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Dash0
OpenTelemetry-native observability with autonomous AI agents that fix production, not just alert on it
ScienceLogic Skylar AI
Agentic AIOps intelligence layer that turns alerts, telemetry and tickets into prioritised advisories and next best actions
A10 AI Gateway
Self-hosted LLM control plane for complexity-aware model routing, per-team token budgets and real-time AI cost governance
Etched Sohu
Transformer-only inference ASIC shipped as rack-scale frontier inference clusters