€5k - €8k
Jobs in the Netherlands, in English.
38,000+ live jobsThesis Internship: Agentic Maintenance
Hybrid in Groningen·Added yesterday (last checked 19 Nov)
Our summary
- Work mode
- Hybrid, Groningen
- Contract
- Internship
- Languages
- Dutch



Skills
Nice to have
This job advert was originally written in Dutch.
Pay & benefits
What you'll get
This is what you will receive from KPN:
- An internship allowance of EUR 600,- (based on 40 hours).
- A laptop, which you can also use for your school assignments.
- Do you not have a student travel pass? Then your travel expenses for the train or car will be reimbursed.
- With us, you will receive a permanent mentor who will coach you professionally during your internship.
- You will get access to a large network of experts at KPN who are happy to share their knowledge & network with you.
- Expand your knowledge & skills through our KPN Academy and workshops.
- Fun activities via Jong KPN and the intern network.
- And also not unimportant; there are regular social drinks and activities at the office.
The role
KPN is the leading telecom and IT service provider in the Netherlands. We manage critical digital infrastructure: from fixed and mobile networks to cloud and platform services. We do this for millions of consumers and businesses. In doing so, we work continuously on reliable, secure, and high-performing services.
You will join the Containerization & Virtualization team within the Networks department. This team builds and manages OpenShift platforms used by multiple customers (tenants) utilizing container technology. Within this assignment, you will research together with the team how recurring maintenance of, for example, code, configuration, and documentation can be handled in a controlled manner by collaborating agents.
As a thesis intern, you will research and develop an agentic workflow through which multiple AI agents can collectively handle simple, recurring maintenance tasks. You will design agent profiles with different roles and build a proof of concept (PoC) in which agents pick up, analyze, execute, test, and controlledly complete user stories. The goal is to demonstrate which maintenance tasks can be automated within clear boundaries and where human intervention remains necessary.
What you'll do
- Select suitable basic maintenance tasks together with the team and translate them into defined user stories with testable acceptance criteria.
- Design a multi-agent workflow with explicit agent roles, task distribution, context transfer, and agreements regarding collaboration and escalation.
- Realize a working PoC with a kanban board, controlled access to tooling via Model Context Protocol (MCP), and human-in-the-loop approval moments.
- Test the workflow on selected maintenance tasks and evaluate and present the quality, lead time, human effort, and limitations.
The central research question is: How can collaborating AI agents, via an agentic workflow, handle basic maintenance tasks from user stories on a kanban board in a controlled and demonstrable manner, using MCP connections and human-in-the-loop at the appropriate moments?
Agent profiles and collaboration
You will set up multiple agent profiles and substantiate their responsibilities, instructions, available context, permitted tools, and authorities. A possible division of roles for the PoC is:
- Coordinator: selects released user stories, distributes the work, monitors dependencies, and updates the board status. Prevents multiple agents from executing the same item simultaneously.
- Analyst/planner: checks whether the user story is executable, gathers relevant context, and drafts an approach including risks, acceptance criteria, and test steps.
- Executor: realizes the defined change in an isolated working environment and submits the change with an explanation for review.
- Reviewer/tester: assesses the change independently of the executor, performs checks, and tests the acceptance criteria. Deviations are sent back for repair or presented to a human.
You will investigate which division of roles actually provides added value. The human supervisor remains responsible for task release, the agreed-upon approval moments, and decisions outside the agents' authorities.
What do you deliver?
- A substantiated selection of suitable maintenance tasks, elaborated as user stories with acceptance criteria. Think of possible examples such as a small dependency update, a simple configuration correction, or updating documentation; you will make the final selection together with the team.
- A technical design including agent profiles, collaboration, MCP connections, access rights, approval moments, and error handling.
- A reproducible, working PoC that handles selected user stories from task intake to controlled completion, including changes, tests, feedback on the board, and human approval.
- An evaluation with pre-agreed measurement criteria: correctly completed tasks, quality of the solution, lead time, required human interventions, tool and model costs, and compliance with safety boundaries. Where possible, compare with manual execution of similar tasks. Also test rejection, missing rights, erroneous input, and failing tests.
- Documentation, a demonstration, and advice on which tasks are suitable for further automation, which limitations remain, and which follow-up steps are required.
The assignment focuses on a proof of concept for a limited set of low-risk maintenance tasks, chosen together with the team, in a non-production environment. The goal is not to automate full platform management or to deliver a production-ready, fully autonomous management solution. Production changes and drastic or irreversible actions fall outside the PoC. You will demonstrate what works within the agreed boundaries, what human control remains necessary, and what constitutes a responsible next step.
Functie-eisen
For this assignment, we are looking for a HBO or WO ICT student with an interest in AI agents, software development, and DevOps automation. Experience with all the techniques used is not necessary. We find it especially important that you are curious, conduct structured research, and can clearly substantiate design choices and results.
Affinity with the following subjects is desired:
- AI agents, language models, and designing agentic workflows.
- Programming, scripting, and connecting tools via APIs and MCP.
- Git, automated testing, and CI/CD.
- Linux, containers, and Kubernetes/OpenShift; Infrastructure as Code and/or GitOps is a plus.
- Secure automation, access management, and human-in-the-loop. Additionally, you command the Dutch and English languages and can clearly document and present your findings.
- You are available for 40 hours per week for at least 20 weeks starting February 2027
- You are in the thesis phase of an HBO or WO degree in the field of ICT.
Hiring process
Are you ready for this challenge and do you want to contribute to the success of KPN? Our belief in technology is great. This becomes even more powerful together with human contact, which is why we would love to meet you!
Want to know more? Contact our recruiter Renée van der Horst via +31 6 49 00 01 20 or renee.vanderhorst@kpn.com. Applications sent by email will not be processed. Please express your interest via the application button.
At KPN, we have recruitment talent that works together smartly and thus wins. Therefore, acquisition in response to this vacancy is not appreciated.
More jobs like this
or browse Groningen·Engineering·Software engineering·Entry-level·Hybrid·AEX·Telecommunications