Software Engineering PhD: Using behaviour specifications and executable tests to check AI coding-agent changes

Software Engineering PhD: Using behaviour specifications and executable tests to check AI coding-agent changes

Quick Decision

Strong but verify fit. This is a timely, clearly motivated software-engineering doctorate: it asks how human-readable behaviour specifications and executable tests can keep increasingly autonomous coding agents aligned with stakeholder intent as requirements and software change. BTH's empirical software-engineering environment is relevant, and the current employer page lists a 15 December 2026 deadline. Before applying, confirm the supervisor team, salary and funding source. As of 5 November 2026, the employer page listed 15 December, but a saved listing also carried an expired marker and the external form was not verified. Confirm the live form and exact cutoff time.

Research Problem and Doctoral Work

What is unsolved and why it matters

Coding agents can generate and modify software, but a patch that passes the current test suite may still fail to reflect what stakeholders meant, especially when requirements are incomplete, ambiguous, conflicting or changing. The proposed project treats testing as a way to keep humans in control: a readable behaviour specification should let stakeholders inspect and discuss expected behaviour, while executable tests provide repeatable checks on generated code and later modifications. The central challenge is to ground both artefacts in requirements and stakeholder intent and preserve a trace from intent to specification, test and code change. The official project description makes these linked artefacts and the human-control problem the core of the work.

The advertised tasks include eliciting and refining expected behaviour from requirements, stakeholder input, developer-agent conversations and development artefacts; identifying gaps and ambiguity with people in the loop; connecting readable specifications to executable tests; maintaining those tests as software evolves; and evaluating effectiveness, reliability and practicality. The role combines method and tool development with experiments and industrial case studies. The exact research questions and methods are to be agreed with the supervision team, so these are project directions rather than a fixed protocol.

Work, methods and potential outputs

A useful early research design would define the unit of traceability (for example, a requirement or stakeholder decision linked to one or more behaviour examples, tests and code changes), choose representative change scenarios, and compare agent behaviour with and without specification-linked tests. Relevant measures could include missed or spurious behaviours, regression detection, specification coverage, test usefulness, trace-link maintenance cost, stakeholder comprehension and the effort required for human review. These are analytic suggestions, not outcomes promised by the vacancy.

Recent work offers helpful baselines without closing the question. TiCoder used tests to help programmers clarify informal intent; a mixed-method study with 15 programmers and additional experiments across four models suggests test-guided clarification can improve evaluation of generated code. Its interaction design is adjacent to this project, but it does not establish that autonomous agents can maintain stakeholder-aligned specifications over long-running repositories. TiCoder's study is a relevant starting point for experimental design.

Agent testing has infrastructure problems of its own. An ISSTA study tested an LLM agent on 50 open-source projects across 14 languages to set up and execute existing test suites, illustrating that repository configuration and test execution cannot be assumed to work uniformly. The cross-project test-execution study can inform environment design and failure categories. An ASE study analysed 120 trajectories and 2,822 interactions from three software-engineering agents; its focus on recurring interaction patterns and failure modes supports evaluating process traces as well as final patches. The agent-trajectory study does not test the advertised specification-and-test intervention, but it illustrates why end-state pass/fail alone may be too narrow.

The most useful first-year conversation is therefore not “which agent is best?” but “which intent and change scenarios can be observed, what counts as a meaningful human-control failure, and what data can the team access?” An industrial case could provide realistic requirements and review practices, but access to a partner, repository or developer-agent trace is not confirmed on the project page.

School, Programme and Field Context

Blekinge Institute of Technology (BTH) is both the employer named in the vacancy and the institution responsible for the doctoral subject. The advert says the post is part of ELLIIT, a multi-university strategic research partnership. Participation in that partnership is not evidence of a joint or double degree; the BTH software-engineering syllabus governs the doctoral subject. BTH reports about 110 active doctoral students across nine subjects. BTH's research-studies overview describes the broader doctoral setting, while the Software Engineering syllabus sets four years of net study (240 credits), including at least 60 course credits and a thesis of at least 150 credits. Each student has an annual individual study plan and at least two supervisors. The position can include up to 20% teaching or other services, with a maximum five-year employment period in the vacancy.

The department's Software Engineering Research and Education Lab Sweden describes an empirical, evidence-based environment with industry collaboration. Its stated themes include AI for Software Engineering, requirements management, software quality, and verification and validation. The official research-area page is a closer match to this project than an overall university league table. BTH also has an ongoing GenAI project with industrial partners that studies coding/documentation workflows, adoption and software quality; its named project manager differs from this vacancy, so it should be read as nearby activity rather than confirmed supervision or shared funding. The GENESIS project page supports the existence of this adjacent research environment.

The ranking evidence is limited. QS lists BTH in a 601-650 subject band for 2021 and 2022; this is dated and does not provide a current project-level measure. BTH's own research overview calls its software-engineering research “second in the world,” but the page does not give a methodology or comparison set. Treat both as background signals, not proof of this particular group's standing. The QS institution profile provides the dated band, and BTH's research overview is the source of the institutional claim.

Supervisor and Lab Assessment

The vacancy says research questions and methods will be developed with a supervision team, but it does not identify the principal or co-supervisors. Nauman Ali is listed as a BTH professor in Software Engineering. His official profile names software testing, software metrics, empirical research and industry collaboration among his interests, which fit the methodological and quality-assurance aspects of the project. His profile is not proof that he has been appointed to supervise this specific post; the official department staff directory should be used to verify who is on the actual team.

Four recent contributions help characterize the local expertise. A 2025 case study by Ali and colleagues used topic modelling and domain-expert input on 5,560 review comments from JabRef to identify recurring quality issues, including testability, maintainability and code structure. It demonstrates empirical software-quality analysis, though it studies one open-source system rather than AI-generated code. The code-review feedback study is the closest in subject to reviewing agent changes. In another 2025 paper, Ali and collaborators proposed a value/feasibility/applicability framework for formulating practically relevant software-engineering questions, initially evaluated in a workshop of 42 senior researchers; the authors note that further evaluation is needed. The Lean Research Inception study is relevant to co-developing a feasible project with industry. A 2026 mapping study by Ali and colleagues reviews 104 papers on operational monitoring data and anomaly detection in cloud systems, while a 2024 preprint maps 174 studies of software analytics across testing, requirements, defects and other tasks. The 2026 monitoring-data paper and the 2024 software-analytics mapping show recent evidence-synthesis and software-quality work; neither is direct evidence of agent-guardrail supervision.

Funding, Application and Living Cost

The employer page lists a 100% temporary doctoral-student position, up to four years of full-time research, and up to 20% other services such as teaching. BTH's policy says doctoral study is financed primarily through doctoral-student employment and that full-programme funding must be secured at admission. BTH's financing rules help interpret the job type, but the vacancy does not give a salary amount, progression scale, pension details or exact funding source. The “not funded by an EU programme” label does not mean the role is unfunded. BTH's general employee-benefits page does not resolve the salary questions for this specific offer.

The current employer page gives a 15 December 2026 application deadline but does not state a closing time or timezone. An earlier listing snapshot supplied 23:59 Europe/Stockholm and also carried an expired marker; that conflicts with the current employer page. On 5 November, the official page displayed an application control and future deadline, but the external form did not load in the web reader. Confirm the form and cutoff directly before submitting. The application list includes a background/experience statement, CV, certified degree certificates and transcripts, contact details for two references (letters are not required at this stage), earliest start date, publication appendix if applicable and a link to the latest/highest-degree thesis.

BTH's 2026/27 fact sheet estimates SEK 10,656 per month for food, accommodation, local transport, phone/internet and miscellaneous expenses. It also says BTH does not provide its own student housing. This is an exchange-student planning figure, not a PhD employee budget or guarantee of housing. Because no salary is published, take-home pay and purchasing power cannot be responsibly estimated. The BTH fact sheet is a starting point for housing and budget questions.

Applicant Fit and Preparation

A technical, mathematical or scientific advanced-level degree is required for the employment/admission route, or equivalent knowledge sufficient for doctoral study. BTH's general admission page also lists a second-cycle degree, 240 credits with at least 60 at second-cycle level, or equivalent knowledge; the Software Engineering syllabus sets subject-specific requirements. The admission page and the syllabus should be checked against the candidate's degree transcripts.

The vacancy requires strong software-engineering and programming skills plus interest in software testing and agentic software engineering. Meriting experience includes software engineering, computer science, AI or data-science master's study; programming, Python, version control and CI/CD; AI coding tools; specification-based or automated testing, behaviour-driven development and requirements traceability; empirical studies and industry collaboration; and strong written and spoken English. These latter items are merits, not all separate hard conditions.

The strongest application evidence would show both implementation skill and research judgement: a reproducible test harness, a short example tracing a requirement into behavioural examples/tests, a code-change evaluation, and clear discussion of failures and limits. An empirical project or industry case would help. These are suggested ways to demonstrate fit, not application requirements.

Next Actions and Questions to Verify

Before applying

1. Confirm that the external form is open and ask for the exact 15 December cutoff time and timezone.
2. Ask who the principal and co-supervisors are, and whether industrial case sites or data access are secured.
3. Request the salary scale, progression, pension, contract and written funding details.
4. Ask which agent frameworks, languages, test infrastructure and repositories will be available at project start.
5. Confirm how English proficiency and certified or translated qualifications are assessed for international applicants.

At interview

Discuss the first-year change scenarios, the baseline agent/test setup, and how the team will evaluate test quality, defect detection, traceability, stakeholder understanding and human oversight. Ask how developer logs and industrial code will be anonymized or protected and what publication restrictions apply. Clarify the expected balance between tool building, experiments, qualitative work and industry case studies.

Before accepting an offer

Obtain the written gross salary, pay progression, pension, benefits, employment duration and confirmed funding source. Confirm supervisor assignment, meeting cadence, industrial access, data-protection rules, housing logistics and the study-plan review process.

Evidence and Sources

Project, institution and training

• Official project and application requirements
• BTH research-studies overview
• Admission to research studies
• Financing of third-cycle studies
• Software Engineering third-cycle syllabus
• Software Engineering research at BTH
• QS institutional profile
• BTH research overview and annual-report figures

Supervisor, lab and recent research

• Department staff directory
• Nauman Ali's official profile
• GENESIS research project
• Supporting quality-issue identification from review feedback
• Towards Lean Research Inception
• Monitoring data for cloud anomaly detection
• Software analytics systematic mapping study
• TiCoder test-driven code-generation study
• Agent test-execution study
• Software-engineering agent trajectory study

Funding, costs and labour-market context

• BTH employee benefits
• BTH 2026/27 living-cost and accommodation estimate
• Swedish Public Employment Service occupational and skills analyses
• Swedish Public Employment Service: occupations with good job prospects
• Statistics Sweden ICT usage in enterprises 2024