Breaking
YC W26 Corelayer joins Y Combinator's Winter 2026 batch FOUNDERS Ex-Goldman Sachs infra engineers Mitch Radhuber & Shipra Jha PITCH An AI on-call engineer for regulated industries CUSTOMERS Finzly · Broadridge · Ridery · Pump · Moda FORBES Named among 21 most promising YC startups YC W26 Corelayer joins Y Combinator's Winter 2026 batch FOUNDERS Ex-Goldman Sachs infra engineers Mitch Radhuber & Shipra Jha PITCH An AI on-call engineer for regulated industries CUSTOMERS Finzly · Broadridge · Ridery · Pump · Moda FORBES Named among 21 most promising YC startups
Company Profile · AI Infrastructure  /  San Francisco

Corelayer wants to take the 3am pager off your engineers

An AI-native production support platform that debugs the way a senior engineer would - by reading the data itself, not just the dashboard. Built by two people who ran pipelines at Goldman Sachs.

Every engineer who has ever carried a pager knows the specific texture of a 3am incident. The phone lights up. A dashboard is red, or worse, everything is green and something is still wrong. You start pulling threads - a recent deploy, a spike in one service, a table that looks off - and you do it half awake, alone, against a clock. Corelayer, a San Francisco startup in Y Combinator's Winter 2026 batch, is building an AI agent to sit that shift instead of you.

The company calls itself an AI-native platform for production software support. In plain terms: it watches your alerts, logs, infrastructure, and the actual data moving through your systems, then uses AI agents to figure out what broke and suggest a fix. The tagline on its site is blunt - "Production on autopilot."

What makes that more than a slogan is where the founders came from. Mitch Radhuber, the CEO, and Shipra Jha, the CTO, both built and ran data infrastructure at Goldman Sachs, on pipelines that moved hundreds of billions of rows a day. Radhuber studied computer science at the University of Michigan and did astrophysics research at Princeton before that. Jha came through Carnegie Mellon and cloud infrastructure work at Oracle. They were, in other words, the people getting paged.

The green-dashboard problem

Here is the observation the whole company rests on. Traditional monitoring tools - Datadog, Splunk, PagerDuty - are very good at watching infrastructure: latency, error rates, CPU, whether a service is up. They are largely blind to the data itself. So a service can report perfect health while a bad value, a missing row, or an unexpected NULL quietly moves downstream and corrupts everything it touches. Corelayer's team describes this failure mode as "slow poison." The system looks fine right up until the business impact shows up somewhere far from the cause.

"The only product I've seen that's been able to catch and fix Heisenbugs." Diego Fernandez · CTO, Ridery

A Heisenbug is a bug that seems to disappear the moment you try to observe it - the kind of ghost that eats a week of engineering time. Catching one is the sort of specific claim that only lands if the product actually does it. Corelayer leans into that specificity rather than the usual broad "AI for DevOps" pitch.

100B+Rows/day the founders ran at Goldman
3People on the team
W26Y Combinator batch
2025Year founded

How it actually works

The reasoning engine has a name: the Production Cortex. Instead of looking at one signal, it stitches together four layers at once - the code repositories (what changed and who changed it), the databases (the real data, not just query speed), the deployments (what shipped and when), and the existing observability tools (logs, traces, metrics). A multi-agent architecture then reasons causally: it traces an anomaly backward through the pipeline, lines it up against recent deploys and infrastructure events, and surfaces root-cause hypotheses in minutes rather than hours.

Code / Repos Databases Deployments Logs / Metrics Production Cortex Root cause Suggested fix
Four signals, one brain. Corelayer's agents read code, data, deploys, and logs together - then reason backward to the cause. Most tools only see the bottom row.

Along the way it does the unglamorous work that makes on-call bearable: filtering false positives, grouping related alerts so one incident does not page five times, and learning a team's systems through feedback. Every agent action comes with an audit trail and citations - not a small detail when your buyer is a bank.

Old way vs. Corelayer

Traditional monitoring
Corelayer
Watches infrastructure metrics
Watches the data itself, plus the metrics
Pages a human to investigate
Agent arrives with a root-cause hypothesis
Green dashboard = "all good"
Catches silent data-quality failures
Cloud-only, one-size-fits-all
On-prem / BYOC, SOC 2, PII masking

Why regulated industries

Corelayer is not chasing every engineering team. It is aimed at finance, healthcare, and insurance - data-heavy, regulated sectors where an outage is expensive and a data leak is worse. Those buyers cannot simply ship customer data to a third party's model. So the product is built compliance-first: on-premises and bring-your-own-cloud deployment, SOC 2, PII masking, confidential compute, and flexible LLM inference. In a category where many competitors are cloud-native by default, that architecture is the wedge.

"I've tried literally every AI SRE product... Corelayer is very impressive." Shri Bhartia · Founding Engineer, Pump

The early customer list reads like the target market talking back: Finzly, Broadridge, Ridery, Rilla, Pump, and Moda - a spread from growth-stage fintech to established financial infrastructure.

FinzlyBroadridgeRideryRillaPumpModa

The neighborhood it competes in

On the incident-response side, the field already has incident.io, FireHydrant, Rootly, and PagerDuty's own AI features. On observability sit the giants, Datadog and Splunk - which are also the real long-term threat, since either could ship a competing feature inside a platform customers already pay for. Corelayer's answer is to plant its flag where the incumbents are weakest: data-quality failures, and a compliance posture that cloud-first tools struggle to match. It integrates with those same tools rather than trying to replace them - reading from Datadog, Splunk, PagerDuty, GitHub, GitLab, Postgres, Snowflake, and the major clouds as inputs.

Where each tool looks (illustrative)
Infra metrics
incumbents
Logs / traces
incumbents
Data quality
Corelayer
Root-cause agent
Corelayer

A rough map, not a benchmark: incumbents own the metrics lanes; Corelayer's bet is the data-quality and autonomous-root-cause lanes.

The business, and the bet

The model is straightforward B2B SaaS: sell an AI on-call and SRE platform to engineering teams, priced against the on-call hours it gives back. Notice that the customer quotes never mention the underlying model or the word "AI" - they talk about saved time and caught bugs. That is the actual product. The larger bet, in the company's own framing, is that the next generation of observability is not about better dashboards. It is about an agent that debugs production the way a senior engineer would.

"Three months ago, I quit my job and moved to SF to build Corelayer. Yesterday, we started the YC W26 batch." Mitch Radhuber · Co-Founder & CEO

It is a three-person company at the start of a batch, with reported seed funding and a Forbes nod as one of W26's more promising names. Whether an agent can really be trusted to root-cause a production incident inside a bank is exactly the question the next year will answer. Corelayer's wager is that the people who have felt the 3am page are motivated enough to find out.

AI SREOn-CallIncident ResponseObservability Data QualityRegulated IndustriesFintechYC W26Developer Tools