How VSCO used RudderStack and Cursor to build an agentic workflow that transformed product tracking and analytics

VSCO is a subscription-based creative platform for photographers and visual creators, offering mobile and desktop tools for photo editing, video editing, collaboration, and portfolio management. As the platform expanded into a broader creative ecosystem, understanding how users engaged across features, devices, and workflows became increasingly important to VSCO.

VSCO’s team initially built their customer data infrastructure in-house, but as the company scaled, cracks started to show. To improve onboarding, measure feature adoption, and support product iteration, VSCO’s team needed fast access to reliable customer data. Their in-house infrastructure made that nearly impossible. But after migrating to RudderStack, they were able to leapfrog traditional workflows completely. Now they’re leveraging agentic workflows with Cursor to overcome tracking challenges and get trustworthy data to their product team at unprecedented speed.

Declaring bankruptcy on homegrown data infrastructure

Every product decision at VSCO starts with a behavioral question. Where do creators abandon their photo editing? Which features separate a hobbyist from a professional? What drives someone to subscribe? For years, VSCO relied on their in-house infrastructure to answer critical product questions, but as the product expanded (more platforms, more teams, more events), it became difficult to keep up.

Before RudderStack, VSCO ran on a homegrown event logging system that demanded constant attention. Schema mismatches, payload errors, and naming inconsistencies, all data quality issues that required manual intervention. The infrastructure was brittle and hard to maintain, and it slowed their team down dramatically. It took five to six weeks from the time a product manager raised a new product question to the time they could have trustworthy data to answer it. "Up until recently, a lot of my time here has been spent maintaining. After a large refactor that didn't have the outcome we wanted, we finally declared bankruptcy and moved over to RudderStack.” says Quijano ‘QJ’ Flores, Staff Data Engineer, VSCO

The Right Way to Build Infrastructure

Initially, VSCO approached RudderStack as a replacement for their existing in-house infrastructure; same workflows, same models, just new tooling. Their RudderStack CSM, Luke Johnson, helped reframe that approach, encouraging the team to build infrastructure designed to scale without constant rework. That shift meant starting with the right foundation from day one.“It wasn’t a migration, it was a paradigm shift,” said Quijano ‘QJ’ Flores, Staff Data Engineer at VSCO. “We stopped trying to replicate our old systems and started using RudderStack the way it was designed to be used. Luke Johnson was instrumental to our success and ultimately set us on a completely different trajectory.”

Three tools made that shift concrete:

• Using RudderStack’s Data Catalog, VSCO defined and centralized definitions of every event and property and standardized them across teams. Before the Data Catalog, engineers instrumented events using language that made sense for their feature at the time. The result: the same user action could have three different names depending on which team had built it first. The Data Catalog fixed this by giving VSCO one place to define every event and property: what it's called, what it means, and how it should be tracked. Product managers, engineers, and data scientists now work from the same vocabulary.
• RuderStack’s Tracking Plans turn that vocabulary into an enforceable contract. For every event in the data catalog, VSCO defined which properties were required, what data types were expected, and what valid values looked like. When an event comes in that doesn't match (a missing property, a wrong type, a value that doesn't conform), RudderStack flags or drops it before it ever reaches the warehouse. The back-and-forth that used to consume weeks ("this event isn't coming through, this property is wrong") gets caught at the source, not discovered downstream.
• RudderTyper reads VSCO's tracking plans and generates native, strongly-typed bindings, ensuring client engineers always include required properties, pass the correct types, and follow existing naming conventions when they instrument tracking. This solved what QJ called one of the hardest problems with their homegrown system: semantic contamination: engineers naming events based on tooling rather than the language used by the product teams.

Together, these RudderStack tools dramatically improved VSCO’s data quality and laid the foundation for everything that followed.

Governance First. Agents Second.

VSCO's first win with RudderStack was trust. The Data Catalog, Tracking plans, and RudderTyper replaced their homegrown system with something the team could rely on. That foundation enabled what came next: an agentic workflow for analytics tracking.

Migrating to RudderStack allowed VSCO to rethink more than their tooling. It gave them the ability to define data ownership and governance for the first time. QJs team adopted data mesh principles and thoroughly documented their data operations to enable teams to own and self-serve their own data products. The result: thirty pages of documentation that was accurate, complete, and structurally sound. But it was so detailed and complex that there was no obvious path to operationalizing it. QJs manager provided feedback: "This is great, but it's too long for a team to get started quickly. Can we get a Quick Start guide?"

Instead of simplifying the framework, QJ converted the 30-page instrumentation doc into a Cursor-based agentic workflow. QJ realized that while not optimized for fast human implementation, the comprehensive documentation was perfect for agents, detailed enough to encode real standards and structured enough for an agent to enforce them. With RudderStack's agent-ready infrastructure under the hood, QJ and his team could empower every PM on the team to self-serve tracking for new features, without sacrificing the data quality they'd spent months building.

A look inside VSCO’s agentic workflow

RudderStack’s code-first capabilities gave VSCO the foundation to operationalize AI agents across their customer data infrastructure. Because every RudderStack capability is accessible through APIs, MCP tools, and CLI commands, humans and agents can build and debug infrastructure, enabling natural language operation across the platform.

This agent-ready infrastructure made it possible for QJ to turn their complex, 30-page tracking doc into an agentic workflow. Product managers can now describe what they’d like to track using natural language in cursor. AI agents then reference tracking plans and the existing codebase to identify instrumentation gaps, generate high-quality tracking code, and automatically generate PRs in GitHub for VSCO’s client engineering team to review and merge. What once required extensive coordination across teams is now implemented in a fraction of the time, dramatically accelerating product tracking.

From Event Tracking to Insight

By putting tracking directly in the hands of product teams, product managers at VSCO can now make decisions in days instead of waiting weeks for instrumentation and validation cycles. Now product teams get the data they need to make decisions when they need it, not after the moment has passed. They’re no longer relying on intuition because they don’t have data. They’re relying on data because they do have it, and they can trust it.

This shift has also allowed the data team to focus on higher-leverage work: building ML models for retention and prediction, conducting deeper funnel and behavioral analyses, and developing agentic tooling to expand data access across VSCO. It started with a solid data foundation: scalable, reliable customer data streaming, and strong, proactive governance. Then RudderStack’s agent-ready infrastructure empowered QJ and his team to innovate for the real transformation.