Doctoral Student (m/f/d): Agentic Modernization of Legacy Enterprise Systems

01.08.2026, Academic staff

Three-year instustrial doctoral student position on agentic modernization of legacy enterprise systems. Employed by AUDI AG in Ingolstadt, supervised at TUM Heilbronn. The research asks how coding agents can transform old COBOL and Java systems in a way that is spec-driven, traceable, and verifiable enough for enterprise governance. Start 01.10. or 01.11.2026, applications to [email protected].

This is an industry doctorate. You will be employed by AUDI AG and supervised academically at TUM, where the doctoral degree will be awarded. The position combines access to production enterprise systems with the freedom and the time to produce peer-reviewed research.

Research focus

Large enterprise systems still run on decades-old code bases whose documentation is stale, whose specifications exist only implicitly in the code, and whose failure has direct business consequences. Generative AI and agentic coding tools promise to accelerate the modernization of such systems, but current practice is largely ad hoc: transformations are neither traceable to a specification nor verifiable against the behavior of the original system, which makes them unusable under enterprise governance constraints.

This position investigates how modernization can instead be made spec-driven, traceable, and verifiable, combining deterministic program analysis with agentic AI components. The work is grounded in real legacy systems, predominantly COBOL and Java, starting from smaller modernization scenarios and progressing toward larger monolithic applications.

Research questions

Your doctoral work will address questions such as:

• How can specifications be recovered from legacy artifacts (source code, tests, configuration, runtime traces, tribal knowledge) when documentation is absent or unreliable? What form should such specifications take to be both machine-actionable and reviewable by humans?
• Which agentic orchestration strategies scale to systems that exceed any context window? How should deterministic analysis (static analysis, type information, dependency graphs, test generation) and LLM components be divided and combined?
• How can traceability be maintained across legacy artifact, recovered specification, and generated target code, such that every transformation decision can be audited after the fact?
• Which verification and governance mechanisms make agentic modernization acceptable in regulated, mission-critical environments? What is the role of differential testing, contracts, and executable specifications?
• How do we evaluate such systems scientifically? What benchmarks, baselines, ablations, and failure-mode taxonomies distinguish generalizable contributions from engineering progress?
• How is domain knowledge embedded in a legacy system preserved rather than silently discarded during modernization?

You will not be expected to answer all of these. Sharpening them into a coherent thesis is part of the work.

Your qualifications

Required

• A very good Master's degree in computer science, software engineering, AI, or a closely related field
• Strong foundations in software architecture, software evolution, testing, and debugging, and the ability to reason systematically about nontrivial systems
• Good programming skills in at least one modern language (e.g., Python, Java, C++, C#, Go, Rust, Kotlin), plus experience with version control, automated testing, and build systems
• Evidence of research potential, through the Master's thesis, research projects, publications, open-source contributions, or technically substantial software projects
• Willingness to design reproducible experiments, benchmarks, baselines, and evaluation metrics, and to engage critically with the scientific literature
• Ability to assess the limitations, failure modes, and reproducibility of AI-assisted software engineering systems critically
• Proficient written and spoken English. German is beneficial but not required.

Advantageous

• Interest in legacy modernization, reverse engineering, and architecture transformation
• Prior exposure to COBOL, mainframes, transaction processing, or enterprise applications
• Hands-on experience with LLMs and AI-assisted software development, including coding agents such as Claude Code, Codex, Cursor, GitHub Copilot, or Gemini CLI
• Experience with APIs, databases, containers, distributed systems, or cloud-native development
• Experience developing large-scale software systems

Candidates are not expected to cover every area above. We particularly welcome applicants with strong depth in one relevant field, such as program analysis, software testing, AI for software engineering, compiler construction, software architecture, or agentic systems, and the ability to expand into adjacent areas during the doctorate.

What we offer

We offer you an exciting and challenging project within a dynamic and collaborative research environment, positioned directly at the intersection of current AI research and real-world enterprise software engineering. Access to production legacy systems, enterprise experts, architects, and governance teams is ensured through our industrial cooperations, giving the work immediate relevance to practice as well as scientific novelty. Employment is with AUDI AG under its collective agreement. Academic supervision at TUM, industrial supervision at Audi, with both sides aligned on the setup.

Location

Primary place of work is Ingolstadt, with regular presence at the TUM chair in Heilbronn.

Applications

Please send a CV, a cover letter explaining your research interests and motivation, a brief summary of previous work experience, relevant certificates and transcripts, and contact details for at least two referees to [email protected]. Applications will be reviewed on a rolling basis until the position is filled.

Kontakt: [email protected]