AnswerThisView only
Shared answer · view only
AIShared by Andrew Ilnicki · Aug 13
LoginSign up
Technology/Can you find papers similar to this and identify research gaps
Can you find papers similar to this and identify research gaps? AgentModernize: Preserving Business Logic in Legacy Modernization with Multi-Agent LLMs and Behavioral Specification Graphs

Literature Survey: Legacy Modernization with Multi-Agent LLMs and Behavioral Specifications

Overview

This survey examines research related to legacy system modernization and identifies key challenges that remain unaddressed by existing work.

Established Research Areas

1. LLM-Based Legacy Modernization

(Generative AI-Driven Legacy System Modernization, 2025) Generative AI enables transformation of monolithic architectures into modern microservices and facilitates code translation, documentation generation, test case creation, and intelligent refactoring recommendations. (Empowering Application Modernization with LLMs, n.d.) Structured frameworks combined with human review processes can modernize legacy applications in contemporary languages like Java and .NET.

However, (Legacy Modernization Using AI Agents, 2026) large language models face significant context limitations—even one-million-token context windows are insufficient for the largest legacy codebases, and tools are largely file- and session-scoped, with understanding lost when sessions end.

2. Multi-Agent Systems for Software Engineering

(AgentMesh, 2025) Software development can require cooperation among multiple members with diverse skills, motivating exploration of agentic frameworks. (Agyn, n.d.) A team-based approach models software engineering as a collaborative process with agents having distinct roles and execution environments. (How Swarms of AI Agents Are, 2026) In practice, multi-agent coordination for debugging workflows achieved a 93% reduction in time-to-root-cause.

3. Program Comprehension and Code Understanding

(Program Comprehension, n.d.) Program comprehension involves top-down and bottom-up processes. (Program Comprehension as a Central Skill, 2025) In the era of generative AI, the critical skill has shifted toward understanding and maintaining AI-generated code. (exa.ai, n.d.) Code understanding challenges include navigation and question-answering.

4. Business Logic Extraction from Legacy Systems

(AgentModernize, n.d.) Business rules in legacy systems are rarely documented—they live in control flow patterns, conditional branches, exception handlers, and configuration files. (Semantic Business Rule Extraction, 2019) Semantic business rule extraction seeks to infer the original intention of procedural logic, complicated by intertwining of technical and business logic. (From COBOL to Models, 2013) Model-driven reverse engineering frameworks extract business logic from legacy COBOL systems.

5. Program Transformation and Semantic Refactoring

(Semantic refactorings, 2017) Syntax-driven refactoring focuses on syntactic transformations, while semantics-driven refactoring uses formal verification to reason over program behavior. (Program synthesis, 2015) Program synthesis can automate refactoring decisions by verifying observational equivalence between original and refactored code.

Critical Research Gaps

Gap 1: Session Persistence and System-Wide Understanding

(Legacy Modernization Using AI Agents, 2026) AI tools operate in session-scoped manner, with understanding lost when sessions end.

Gap 2: Behavioral Correctness During Transformation

(AgentModernize, n.d.) A syntactically correct translation can pass compilation but silently break edge-case handling, alter validation logic, or drop constraints that were never written down.

Gap 3: Behavioral Specification Extraction

(AgentModernize, n.d.) Silent behavioral changes during migration remain a central risk when business rules are not documented.

Gap 4: Large-Scale Codebase Context

(Legacy Modernization Using AI Agents, 2026) Even one-million-token context windows are insufficient for the largest legacy systems spanning hundreds of thousands of files.

Gap 5: API Integration During Code Translation

([2603.14054v1] LegacyTranslate, n.d.) Direct application of LLMs for code translation often produces syntactically valid outputs that fail to compile or integrate within existing production frameworks.

Gap 6: Intermediate Representations for Modernization

[(Software Transformation, n.d.) Program transformation literature discusses representations ranging from source code to intermediate representations to specification.

Conclusion

The surveyed literature shows mature individual techniques for legacy modernization, multi-agent orchestration, program understanding, business logic extraction, and semantic refactoring. Significant challenges remain in integrating these capabilities into systems that maintain behavioral correctness while scaling to large codebases.

Save

Citation Lists

Unknown author · 2026 · internet

Unknown author · 2025 · internet

Unknown author · 2015 · internet

Suggested next steps

Run a sensitivity analysisTurn this into a reportAnalyze an uploaded PDFEdit the draftExport the extraction table
TRY:
History

Loading sources…

Q1

Can you find papers similar to this and identify research gaps? AgentModernize: Preserving Business Logic in Legacy Modernization with Multi-Agent LLMs and Behavioral Specification Graphs

Abstract
Relevant extracts
Source 1: How Swarms of AI Agents Are Redefining Software Engineering

2026 · internet

At the core of the agentic engineering is a collaborative system of intelligent agents designed to mirror how engineering teams plan, execute, and deliver software. Rather than treating AI as a collection of isolated assistants, this framework models agents as team members— each with defined responsibilities, shared context, and accountability— coordinated through a lightweight but powerful leader…

  • What is agentic engineering? Agentic engineering is a multi-agent coordination model where AI agents act as digital team members — each with defined roles, shared memory, and a common observability layer — to move software through the full delivery pipeline, not just generate code faster.
  • What results can multi-agent systems produce in software delivery? In a pilot of 20+ debugging workflows,…
Source 2: Generative AI-Driven Legacy System Modernization: Transforming Enterprise Infrastructure Through Automated Code Translation and Refactoring | Journal of Computer Science and Technology Studies

2025 · internet

Not available

Legacy systems continue to form the operational backbone of numerous enterprises despite presenting significant challenges, including scalability constraints, integration limitations, and escalating maintenance costs. Generative artificial intelligence, particularly large language models trained on extensive code repositories, offers unprecedented capabilities for automating critical aspects of le…

Source 3: Program synthesis: challenges and opportunities

2015 · internet

Not available

During the development process, very frequently situations arise where long-term code quality is traded for short-term gain. Under the pressure of deadlines, developers find quick solutions for fixing bugs or adding support for new functionality which result in code that is too complex, bloated by duplication and hard to understand. This leads to what is known as technical debt, a metaphor introdu…

Source 4: Empowering Application Modernization with LLMs: Addressing Core Challenges in Reliability, Security, and Quality

internet

Not available

Instead, this paper focuses on modernization efforts for systems developed in more contemporary languages, such as Java and .NET, hereafter referred to as legacy applications. These systems are well suited for leveraging LLMs, given the abundance of available training data and the demonstrated proficiency of LLMs with recent technologies. Additionally, as these applications often operate across th…

Source 5: AgentModernize: Preserving Business Logic in Legacy Modernization with Multi-Agent LLMs and Behavioral Specification Graphs

internet

A Specification Generator that encodes the extracted logic into Behavioral Specification Graphs (BSGs) — a structured intermediate representation with preconditions, postconditions, and invariants. ... (KDM) [24] and Architecture-Driven Modernization (ADM) standards from OMG provide formal representations of legacy system semantics for modernization tooling. Behavioral Interface Specification Lang…

The obvious response is modernization. The less obvious problem is that most modernization efforts destroy the very thing they should preserve: behavioral semantics. Business rules in legacy systems are rarely documented. They live in control flow patterns, conditional branches, exception handlers, and configuration files [3]. A line-by-line COBOL-to-Java translation can compile and run yet silent…

Source 6: Program Comprehension as a Central Skill in CS Education in the Era of ...

Yael Erez · 2025 · internet

The answers to these questions depend on the student’s ability to comprehend code deeply. As students increasingly rely on AI to generate their code, without strong comprehension skills, they will not be able to distinguish between correct AI-generated code and AI code that is dangerously flawed. Computer science educators should consider shifting some of the focus from code production to code com…

In the era of generative AI (GenAI), in which AI code assistant tools handle much of the code production, the critical skill that sets competent programmers apart has shifted from writing code to understanding, evaluating, and maintaining AI-generated code, making program comprehension an essential component of computer science education. The central question for learners and professionals is no l…

Source 7: From COBOL to Models: an MDE framework to extract business logic out of legacy COBOL systems

Jordi Cabot · 2013 · internet

Continuing with our reverse engineering ... line, today we present our ... for their daily operations. This ... implements the business rules in place ... , which must be continuously adapted in response to market ... . This evolution implies understanding ... evolving also the underlying software components enforcing those rules. ... This is challenging because, first, the code implementing the r…

This is challenging because, first, the code implementing the rules is scattered throughout the whole system and, second, most of the time documentation is poor and out-of-date. This is especially true for older systems that have been maintained and evolved for several years (even decades). In those systems, it is not even clear which business rules are enforced nor whether rules are still consist…

Source 8: Semantic Business Rule Extraction

2019 · internet

Not available

The process of semantic business rule extraction from legacy code, which seeks to extract the meaning of the rules from the legacy code, always is performed on a procedural language. It is a process of inferring what the data means and what was the original intention of the procedural logic. This is complicated by the almost universal intertwining of technical logic (sending/receiving messages, re…

Source 9: Agyn: A Multi-Agent System for Team-Based Autonomous Software Engineering

internet

To mirror real-world development practices, the system executes a GitHub-native workflow that closely resembles human collaboration. Issues are analyzed, changes are proposed through pull requests, feedback is provided via inline code reviews, and revisions are iteratively applied until the solution is accepted. The entire process is fully automated and does not involve any human intervention. To…

Beyond aggregate performance, the central contribution of this work is demonstrating that team-based coordination among specialized agents is a viable and effective paradigm for autonomous software engineering. Rather than relying on a single monolithic agent or a rigid execution pipeline, our system models development as a collaborative process involving agents with distinct roles, responsibiliti…

Source 10: https://exa.ai/library/publication/bqc0ftd8b9m

Alex Gu, Naman Jain, +5 · internet

Not available

One important challenge in code understanding is code navigation: finding where relevant func tionality is implemented. Doing this well requires understanding the high-level layout of where every functionality lies in the codebase and the low-level understanding of which helper functions are used to implement each functionality. Another challenge is code question answering: answering complex quest…

16 sources · 0 selected
1 / 2