Simulithic, fresh out of Y Combinator, thinks it has cracked a problem that has nagged engineering teams for years: how to catch bugs in production without writing mountains of test scripts that break the moment someone redesigns a button.
The San Francisco startup began letting early users in late September, according to posts from co-founder Satyam Singh. Instead of asking engineers to script synthetic tests that simulate user behavior, Simulithic watches real users, trains AI agents on what it sees, and then runs those learned patterns across browsers and devices every half hour.
It's an approach that could appeal to companies weary of maintaining brittle test suites. Traditional synthetic monitoring demands someone write code that clicks through checkout flows or login sequences, then update that code whenever the interface shifts. Simulithic sidesteps the scripting altogether.
The company's Behavior SDK sits in production, quietly recording how users move through an application: which routes they take, what actions they perform, how long things take, what devices they're on. According to product documentation, the SDK deliberately avoids collecting screenshots, DOM elements, text inputs, field values, or anything personally identifiable.
Engineers type a single command—simulithic map—and the system builds a representation of actual user journeys. AI agents then replay those flows across Safari, Chrome, and Firefox on macOS and Windows, plus mobile browsers on iOS and Android under varying network conditions. When something breaks, the team gets a Slack or email alert with a session replay pinpointing the failure.
Learning from Reality
The pitch centers on one assertion: real user behavior is more valuable than anything an engineer can anticipate in a script. Singh and co-founder Ryan Ning, who serves as CEO, came to that conclusion after stints at places where deployment anxiety ran high. Ning previously engineered software at Uber and Shopify and worked on AI research at the University of Toronto, his LinkedIn profile shows. Singh's resume includes time at Ramp and Bree, a fintech startup from Y Combinator's Summer 2021 batch.
"We realized we were monitoring production every single time I push a change," Singh wrote in his October announcement.
The two founded Simulithic as part of Y Combinator's Fall 2026 batch.
Pricing Stays Under Wraps
Simulithic offers three tiers, though it declined to share dollar figures publicly. The Starter plan monitors three flows every four hours, handles 2,000 web sessions monthly, and includes 30 pull-request checks. Team accounts bump that to 10 flows checked hourly, 8,000 monthly sessions (tripled for mobile traffic), and unlimited PR checks on altered flows. Enterprise customers can monitor 30 flows at any cadence, process 25,000 sessions, retain replays for 90 days, and add single sign-on plus workspace segmentation.
The product also hooks into CI/CD pipelines. When a developer opens a pull request, Simulithic runs the same simulated users through the new code, then drops a verdict per flow and device directly into the PR comments.
Competing with Giants
Simulithic enters a crowded field. Datadog Synthetic Monitoring lets engineers build code-free browser tests, but still relies on step-by-step assertions someone has to define. New Relic Synthetics uses JavaScript-based scripted monitors. Grafana Cloud's k6 Synthetic Monitoring and Checkly both lean on Playwright checks that require writing and maintenance.
The incumbents have scale and brand recognition. What Simulithic offers is a shortcut: skip the scripting, let the AI learn from production traffic, and see if that approach can keep pace with the way modern software actually gets used. It's a bet that learning from reality beats predicting it.
The CLI tool went live on npm at version 0.2.0, the package index confirms. Whether AI-driven monitoring proves more resilient than human-written scripts will depend on how well those agents can adapt when interfaces change in ways real users haven't yet encountered. For now, early adopters willing to let an SDK watch their production traffic can find out.
