Anand
Kulkarni

The 50-Year-Old Code Running Your Life

The COBOL that processes your paycheck was written before Anand Kulkarni was born. The mainframe software clearing your credit card transaction predates the internet. The code adjudicating your insurance claim has had more developers cycle through it than anyone can count - each one leaving behind their own logic, their own patches, their own undocumented workarounds. The original architects? Long retired. The documentation? Never existed. The risk of touching it? Enormous.

This is not a niche technical problem. It is the hidden foundation of global commerce, government services, and critical infrastructure. Trillions of lines of legacy code sit inside the world's most important institutions, written decades ago, understood by no one, impossible to safely modify - and yet running everything.

Kulkarni built a company to read it.

CoreStory, the Berkeley-based AI company Kulkarni leads, is not attempting incremental improvement on an established workflow. It is attempting to replace 18-month expert review processes with something that takes minutes. The platform's AI can analyze 100,000 lines of code - documenting business rules, system architecture, dependencies, and hidden logic - in the time it takes a human team to read the first module.

The results, as reported by enterprises already using the platform: up to 50% reduction in human development time on modernization projects. Not because the AI is writing better code. Because for the first time, the teams actually know what the existing code does.

From Polytope Geometry to COBOL Archaeology

The arc of Kulkarni's career follows an unusual logic: from abstract mathematics to human computation to AI development platforms to enterprise code intelligence. Each step looks like a pivot. From the inside, they form a continuous thread about systems, information, and scale.

At UC Berkeley, Kulkarni was a National Science Foundation graduate research fellow studying polytope theory - the mathematics of high-dimensional shapes, best known to general audiences through the Hirsch conjecture, a longstanding open question about the diameter of convex polytopes. He published more than a dozen peer-reviewed papers across ACM, AAAI, and IEEE publications. His early projects as an undergraduate won prizes at the Mathematical Contest in Modeling and the Berkeley Technology Breakthrough Competition. He was a National Merit Scholar at 17.

None of that is a straight line to enterprise AI. But in the middle of his PhD work at Berkeley, Kulkarni got interested in crowdsourcing - in what happens when you route human intelligence at scale through computational systems. That interest became LeadGenius, a Y Combinator S11 startup he co-founded and served as Chief Scientist, combining human computation with deep learning to automate account-based marketing. LeadGenius raised over $20 million and counted Google, eBay, and Box among its customers.

From LeadGenius, Kulkarni founded Crowdbotics in 2017 - an AI-powered rapid application development platform that served Fortune 500 companies and U.S. government agencies, growing to $20.7 million in annual revenue and 175 employees by 2024, with 200-300% annual revenue growth for three consecutive years.

The pivot to CoreStory wasn't a reinvention. It was Kulkarni looking at the customer problems Crowdbotics kept encountering - complex legacy systems that blocked modernization before it could even begin - and deciding to attack the root cause directly.

What Specification-Driven Development Actually Means

CoreStory's core intellectual contribution is what Kulkarni calls Specification-Driven Development. The concept: instead of working directly with source code, you first convert that code into structured specifications - detailed, queryable documents that capture business rules, workflows, dependencies, and architectural decisions. The specs become the ground truth. Changes are made against the specs. The code is generated or modified from there.

This sounds obvious until you realize that for most legacy systems, no specification ever existed. The code IS the spec. The business logic lives inside 1970s COBOL subroutines written by people who have been retired for two decades. The only way to understand what the software does is to read every line - which, for systems running hundreds of millions of lines, nobody actually does in full.

CoreStory's AI platform attacks this with a technique called recursive decomposition and recomposition. The platform breaks down the codebase into hierarchical layers - modules, functions, logic branches - analyzes each layer, captures what it does, then reassembles a complete intelligence model that includes business requirements, rules, workflows, technical dependencies, and architectural insights. The result is an "Intelligence Model" that functions as the specification the original code never had.

The Intelligence Model doesn't sit in a document. It is a living layer that integrates with the tools developers already use - Claude Code, Cursor, GitHub Copilot - via MCP and API connections. Every developer on the team, every AI coding agent working in the system, has access to the same ground-truth understanding of what the code does and why.

The platform serves four primary use cases: legacy app modernization, application maintenance, AI-generated code management, and developer onboarding. The common thread: understanding complex systems that would otherwise require years of institutional knowledge to navigate safely.

Who Bet $32 Million - and Why

In October 2025, CoreStory closed a $32 million Series A. The round was co-led by Tribeca Venture Partners, NEA, and SineWave Ventures, with additional participation from Harrison Metal, Singtel Innov8, Samsung Next, Nimble Partners, and Alumni Ventures.

The investor thesis is not complicated: legacy software is one of the largest unsolved problems in enterprise technology. Estimates put the total volume of legacy code running critical global systems in the trillions of lines. The annual cost of maintaining systems that nobody fully understands runs into hundreds of billions of dollars. Every attempt to modernize runs the risk of breaking systems that are, by definition, not well-understood.

CoreStory's board also includes representation from Jackson Square Ventures, Homebrew, and Bee Partners - investors who backed the company's earlier work at Crowdbotics and followed it through the rebrand and repositioning. The continuity of investor support through a major strategic pivot is its own signal about confidence in Kulkarni's execution.

The total capital raised across the Crowdbotics-to-CoreStory journey stands at over $50 million. The Series A specifically funds scaling CoreStory's enterprise go-to-market efforts and deepening the platform's AI capabilities.

The Full Arc

Things That Don't Fit Elsewhere

Playing the Long Game

There is a particular kind of founder who builds for the long haul rather than the next funding cycle. Kulkarni has been explicit about this. He has publicly pushed back against the tendency to optimize for twelve-month hype windows rather than ten-year relevance. It is the kind of thing founders say; in his case, it tracks with behavior.

Crowdbotics, the company CoreStory grew from, spent years building deep relationships with government agencies and large enterprises - customers who move slowly, require extensive trust-building, and rarely end up in tech press features. The 200-300% annual revenue growth that resulted was not the outcome of viral adoption. It was methodical expansion into the kinds of complex, high-stakes software environments that require exactly the kind of patient, rigorous approach Kulkarni brings from his academic background.

The same long-term instinct shows up in the CoreStory thesis. Legacy software modernization is not a trend. It is not driven by a new paradigm in venture preference or a recent shift in enterprise buying behavior. It is a problem that has existed for fifty years and will exist for fifty more. The banks and governments running on 1970s COBOL are not going to solve that problem this year, or in five years. They are going to solve it decade by decade, one system at a time.

Kulkarni is building for that timeline. The CoreStory platform is explicitly designed not as a one-shot modernization tool but as a persistent intelligence layer - something that lives in the development workflow, updates as code changes, and serves as institutional memory that doesn't retire when a senior engineer does.

The bet is that the world's most mission-critical software is going to need to be understood for as long as it continues to run. And it will keep running longer than anyone expects.