Large-scale enterprise software systems grow harder to maintain and modernize every year. Their complexity, incomplete documentation, and the gradual loss of institutional knowledge all add to the burden. Recent advances in large language models (LLMs) and AI coding agents have significantly improved software development productivity. Ensuring the correctness, completeness, and trustworthiness of AI-driven software transformation remains a fundamental challenge.

In this talk, Haifeng Chen will present research on AI-powered software engineering assurance, a new paradigm that goes beyond code generation to recover, document, and preserve the underlying business intent of existing software systems. He will introduce Codocs, an agentic AI framework that combines structural code analysis, automated documentation generation, execution-based validation, and iterative refinement to produce trustworthy, verifiable representations of complex software behavior.

The talk will cover key research challenges in scaling to large code repositories, managing interdependent components, and optimizing agentic reasoning through multi-frontier tree search. Haifeng will close by demonstrating how this methodology enables reliable, efficient enterprise software transformation through use cases in software modernization, implementation validation, and compliance auditing.