PhD Measuring and Scaling Agentic Software Engineering (f/m/x)

BMW AG; München; On-site; EUR 36,000 - 47,000; Part time

Benefits offered by this job

Mentoring & onboarding

Flexible working hours

Digital offers & mobile working

Annual special payments

Apartment offers for students (subject

Job summary

BMW Group’s ProMotion PhD programme bridges doctoral studies with real-world mobility challenges. You will design benchmarks for coding agents on automotive codebases, investigate loss points between exploration and generation, and develop methods to preserve and reuse relevant code.

You will also build verification approaches to keep agent-generated code trustworthy without slowing development and benchmark configurations across embedded C/C++, Java, Python and Go.

Qualifications

• Studies in computer science, data science, mathematics or a comparable field.
• Strong knowledge of machine learning, especially large language models and agentic systems.
• Solid programming skills in Python; experience with C/C++, Java or Go is an advantage.
• Hands-on experience with agent evaluation and benchmarking (e.g., SWE-bench, Terminal-Bench, tau-bench).
• Ideally first publications or submissions at relevant scientific venues during your studies.
• Very good English language skills; German is an advantage.

Responsibilities

• You design benchmarks that evaluate coding agents on real automotive codebases across the full development lifecycle.
• Investigate where agents lose relevant code between exploration and generation and how to prevent it.
• Explore methods to help agents preserve and use relevant code more effectively.
• Develop verification and testing approaches to keep agent-generated code trustworthy without slowing development.
• Benchmark agent configurations, LLMs, harnesses and scaffolding across stacks like embedded C/C++, Java, Python and Go.
• Translate findings into research and deployment frameworks for real-world scale.

Skills

Machine learning

Large language models

Agentic systems

Python

C/C++

Java

Go

Benchmarking

Education

Studies in computer science, data science, mathematics or a comparable field

Tools

SWE-bench

Terminal-Bench

tau-bench

Job description

THE FUTURE IS WHAT WE CREATE WHEN WE TAKE OUR VISIONS TO THE LIMITS.

PHD AT THE BMW GROUP.

Bridging the gap between your doctoral studies and the kind of career challenge you’re looking for isn’t always easy. That’s why our ProMotion PhD programme gives you the opportunity to apply your passion and scientific expertise to very real challenges that could shape the future of mobility. If selected, you’ll have the chance to really develop your own future career path with us through your practice-related thesis, promote your chosen specialty from within – and position yourself to secure the career you’ve always wanted.

Our New Technologies team explores and evaluates emerging technologies and brings the most promising ones into productive use across the BMW Group. Agentic software engineering is one of these technologies, and as a PhD candidate, you help find ways to systematically measure agentic software development and scale it to real-world use.

What awaits you?

You design benchmarks that evaluate coding agents on real automotive codebases across the full development lifecycle.

Within this project, you investigate where agents lose relevant code between exploration and generation - a failure mode that grows with codebase size and remains unmeasured.

Building on these insights, you explore methods that help agents preserve and use relevant code more effectively.

In this role, you help develop verification and testing approaches that keep agent-generated code trustworthy without slowing down development.

Together with the team, you benchmark agent configurations, LLMs, harnesses and scaffolding across heterogeneous stacks such as embedded C/C++, Java, Python and Go.

Finally, you translate your findings into research and deployment frameworks that can be applied at real-world scale.

What should you bring along?

Studies in computer science, data science, mathematics or a comparable field.

Solid knowledge of machine learning, in particular large language models and agentic systems.

Strong programming skills in Python; experience with C/C++, Java or Go is an advantage.

Hands‑on experience with agent evaluation and benchmarking, for example SWE-bench, Terminal-Bench, tau-bench or comparable frameworks.

Ideally first publications or submissions at relevant scientific venues during your studies.

Very good English language skills; German is an advantage.

What do we offer?

• Comprehensive mentoring & onboarding.
• Personal & professional development.
• Flexible working hours.
• Digital offers & mobile working.
• Attractive & fair remuneration.
• Annual special payments such as vacation pay and Christmas bonus.
• Apartment offers for students (subject to availability & only Munich).

Start date: 02.11.2026

Duration: 36 months

Working hours: Part-time

You can find helpful tips on your application and the application processhere .

Note: Citizens of countries outside the European Union must have a valid residence or work permit for the duration of the program.

At the BMW Group, we place great importance on equal treatment and equal opportunities. Our recruiting decisions are based on the personality, experience, and skills of the applicants. Learn morehere .