Artificial intelligence startup CoreStory Inc. today announced that it has closed a $32 million funding early-stage round led by Tribeca Venture Partners, NEA and SineWave Ventures.

The Series A deal also drew contributions from several other backers. The group included Samsung Next, the venture capital arm of Singaporean internet provider Singtel, Harrison Metal, Nimble Partners and Alumni Ventures.

Many large enterprises have decades-old applications written in legacy programming languages such as COBOL. Modernizing those applications often requires rewriting them in a new language. Before developers can begin rewriting a program, they have to create documentation that explains what each of the program’s components does and how.

Fully documenting an application can take months in some cases. One reason is that legacy software is often written in COBOL, a programming language that relatively few developers understand. Another factor is that some legacy applications include tens of thousands of lines of code or more. 

San Francisco-based CoreStory is working to speed up the software modernization workflow. The company has developed an artificial intelligence platform, Code-to-Spec, that can automatically generate documentation for legacy code bases. It also provides pointers on how developers should go about modernizing the code.

After analyzing a legacy application, Code-to-Spec generates a natural language overview of its features. It also creates a summary of the technologies that power the program. According to CoreStory, Code-to-Spec maps out details such as the programming languages in which a workload is written and the database it uses to store information.

Alongside high-level architectural summaries, the platform can generate an explainer of each individual code snippet in an application. The explainer comprises two elements. There’s a natural language description, as well as a graph that visualizes the processing steps a given code snippet performs. 

A chatbot interface enables developers to ask Code-to-Spec questions about an application’s code. A user could, for example, request a list of components that could be made more hardware-efficient. Code-to-Spec can rank the code snippets it finds based on the amount of work it would take to rewrite them. 

The platform not only identifies application components that can be modernized but also provides advice on how to go about the task. For example, Code-to-Spec might suggest that developers equip a code snippet with an error handling mechanism to prevent outages.

CoreStory says its platform is powered by model ensembles. A model ensemble is a collection of neural networks that generates multiple answers to each prompt and then combines them into a single response. In some cases, that arrangement can provide higher output quality than other machine learning approaches.

““Every time you swipe a card or make a claim, you’re depending on software written decades ago that few people understand,” said Chief Executive Officer Anand Kulkarni. “CoreStory’s agentic AI reads, maps and documents that code, turning it into living specifications that make modernization safe and measurable.”

The new funding should put CoreStory in a better position to compete with products such as Anthropic PBC’s Claude Code. The programming assistant, which is generating more than $500 million in annualized revenue, lends itself to application modernization projects. Claude Code can rewrite software in a new language and automatically run tests to ensure the code works as expected. 

Photo: Unsplash