PROFES 2026: Product-Focused Software Process Improvement (Karlskrona)

1. Logistics
2. Tracks and calls
- 2.1. Workshop and tutorial
3. Keynotes
4. Papers to see
- 4.1. When Coding Outpaces Understanding: Making Cognitive Debt Observable in AI-driven SE
- 4.2. Others on generative AI, testing and governance of agent-written code
5. Where our research meets it
6. Notes

1. Logistics

Event: PROFES 2026, the 27th International Conference on Product-Focused Software Process Improvement
When: Monday 2026-11-30 to Wednesday 2026-12-02. Monday is workshop, tutorial and doctoral symposium day; the main conference runs Tuesday and Wednesday.
Where: Blekinge Tekniska Högskola (Blekinge Institute of Technology, BTH), Karlskrona, Sweden. In person: accepted papers must be presented in person by at least one author.
Social: reception and dinner Monday evening; Karlskrona guided tour and dinner Tuesday
Link: Home page, Programme

2. Tracks and calls

All papers in Springer LNCS format, in English, as PDF through EasyChair. Main-track deadlines were 3 August 2026 (extended from 23 July), notification 11 September, camera-ready 25 September.

Track	Length	Notes
Research papers	16 pp including references	single-anonymous, at least three reviewers
Industry papers	16 pp; or a 2-page extended abstract for a talk	talk abstracts are not in the proceedings
Short papers	8 pp including references
Posters	6 pp extended abstract	problem, approach, results so far
Doctoral symposium	6 pp PhD proposal; 2 pp for master students	advisor letter of support; master abstracts unpublished
QUEMALES workshop	12 pp full, 8 pp short	submission 7 Aug, notification 4 Sep

The Research Papers call lists twenty topics. Grouped:

process modelling, management, assessment, measurement and metrics, including in AI-augmented settings;
product quality models for software, data, AI, systems and services;
agile processes (scaled, hybrid, remote) and infrastructure for continuous delivery and deployment;
integration of generative AI into SE workflows, and processes for software and AI co-engineering;
modernisation and lifecycle management, open-source development and software supply chains;
human-centred processes, education and training, sustainability;
trustworthy and responsible systems; compliance, governance and regulation.

The call also welcomes industry-academia collaborations, case studies, experience reports and secondary studies that synthesise findings.

2.1. Workshop and tutorial

QUEMALES 2026: Quality Evaluation of ML-based Software Systems. Quality attributes, metrics and measurement frameworks for ML-based software, as a step towards certification; topics include training-data quality, certification and quality cases, and ISO/IEC 25059. Monday afternoon, Lecture hall J1280.
ASCISE: Applied Statistical Causal Inference in Software Engineering, a tutorial by Julian Frattini, Hans-Martin Heyn and Roberto Pietrantuono. Monday morning, two 90-minute sessions, Lecture hall J1280.

3. Keynotes

Tue 1 Dec, 09:00: Darja Šmite, The Return of the Superhero Developer?
Wed 2 Dec, 09:00: Markus Borg, Storm Shelters: Weathering the Agentic Hurricane with Processes and Governance

Both in Multisalen Hall J1504. The doctoral symposium has its own keynote on Monday: Michael Unterkalmsteiner, Uncertainty is the Job: Navigating the PhD Journey.

4. Papers to see

4.1. When Coding Outpaces Understanding: Making Cognitive Debt Observable in AI-driven SE

Track: Short Papers and Posters (position paper)
Authors: Jefferson Seide Molléri (Kristiania University of Applied Sciences); Antonio Martini (University of Oslo, Norway)
Slot: Tue 1 Dec, 16:45 to 17:00, Graduation Hall J1620, Session 3b: Human-AI Collaboration, Adoption, and Cognitive Debt
Link: Paper page

The abstract argues that AI-driven software engineering may increase cognitive debt through an AI-induced comprehension gap: teams produce artifacts faster than they understand them. It proposes a conceptual model linking the gap to acceleration mechanisms and observable indicators, outlines research directions for making cognitive debt observable, and argues that software process improvement should treat understanding as a deliverable rather than a by-product of implementation.

Authors' related work (checked 2026-11-13; arXiv entries against the arXiv API, the rest against OpenAlex/Crossref DOIs; the paper itself has no preprint or artefact yet, so its indicators are not known before the proceedings):

Jefferson Seide Molléri (Kristiania), empirical-SE method: "Who's actually being studied?" (WSESE 2024, https://doi.org/10.1145/3643664.3648200); "Core view of research quality in ESE" with Mendes, Petersen and Felderer (CSI 2022, https://doi.org/10.1016/j.csi.2022.103688); "Plan4code" with Justo and Martins (ICSE Companion 2026, https://doi.org/10.1145/3774748.3787651).
Antonio Martini (Oslo), technical debt: "Process Debt: Definition, Risks, and Management" with Stray, Besker, Moe and Bosch (JSEP 2025, https://doi.org/10.1002/smr.70017), the nearest earlier debt that is not in the code; "TD Friction for Maintenance Prioritization" with Borg, Tornhill and others (ICSME 2026, https://doi.org/10.1109/icsme67971.2026.00120); LLM refactoring of architectural smells (EMSE 2026, https://doi.org/10.1007/s10664-026-10939-8).
His group's repositories (via Karthik Shivashankar): debtbuster, an LLM-agent harness that audits and refactors Python and gates it through three checks that use no LLM (MIT); PyExamine (MSR 2025) and MLScent (CAIN 2025), smell detectors.

Closest earlier work on cognitive debt:

Storey, From Technical Debt to Cognitive and Intent Debt (arXiv, 2026), a model of technical, cognitive and intent debt; and Cognitive Tours: Making Understanding a Deliverable (arXiv, Nov 2026), nearly the paper's own slogan.
Tavantzis, Trinkenreich, Greiler et al., Cognitive Debt in the GenAI Era (arXiv, 2026): 76 practitioner sources; the debt hides behind passing tests.
Borg et al., Echoes of AI (arXiv, 2025), downstream maintainability; Kosmyna et al., Your Brain on ChatGPT (arXiv, 2025), an EEG study of essay writing, not software.

Where it meets our notes: The oracle is the deliverable parallels "understanding as a deliverable"; Harness first's "green can mean never tried" is Tavantzis et al.'s debt behind passing tests, and its reach counts are an observable indicator of the kind the paper asks for. Inference, not stated by either side: the elenctic review could serve as a comprehension gate, and the hardening triggers could double as points to check cognitive debt.

4.2. Others on generative AI, testing and governance of agent-written code

From the published programme. Times are start times; J1610 and J1620 are Graduation Halls.

Monday 30 November (QUEMALES, J1280)

15:05: Michel Nass, Semantic Mission Invariance Testing for LLM-Based Drone Agents
16:00: Bogdan Andrei, Ana Oprescu, Sander Klous, A Survey of Quality Characteristics for Code-Generating Large Language Models
16:25: Liang Yu, Jon Påhls, Emil Alégroth, Panagiota Chatzipetrou, Measuring the Quality of LLM Code-Generation Systems

Tuesday 1 December

10:50, J1610 (industry)

Bexell, Gupta, Gullstrand Heander, Söderberg, Runeson, Eldh, Shi, Malysh, Helpful but Fallible: Developer Experiences of AI Tools Under a Coordinated Industrial Rollout

13:30, J1610 (research)

Miroslaw Staron, Who Changed the AI Agent's Rules? Measuring How AI Coding Agents Escalate Authority in CI/CD Pipelines

13:50, J1610 (research)

Laiq, Britto, Usman, Saini, Badampudi, Using Agentic AI for contextualized and multifaceted code review at Ericsson

14:30, J1610 (research)

Siddeeq, Waseem, Malhi, Abrahamsson, CodeAssay: A Multi-Metric Benchmark with Audited Ground Truth for LLM Code Generation

14:30, J1620 (industry)

Orucevic-Alagic, Alemo, Orheim, Runeson, Vucic, CAFE: A Context-Aware Framework for Selecting Third-Party Software Tools – An Industrial Case Study of Web API Fuzzing

15:30 to 17:00, J1610

the whole of Session 3a, Generative AI for Software Testing:

Zhao, Tanveer, Zdravkovic, How GenAI Reshapes Software Testing: Changes in Process and Roles (industry)
Jansson, Strandberg, Sörensen, Enoiu, Afzal, Supporting Industrial Test-Failure Analysis with LLM-Based Systems: An Experience Report (industry)
Dasanayake, Mohanani, Mandić, Turhan, Generative AI in Software Testing: Practitioners' Perspectives on the Benefits, Challenges, Opportunities, and Risks (short)
Ashraf, Sami, Rasku, Niemistö, Abrahamsson, From Pixels to Structured Interaction: An Industrial Study of MCP-Driven LLM Agents for UI Test Automation (industry)

16:25, J1620 (industry)

Lakmal Silva, Michael Unterkalmsteiner, Krzysztof Wnuk, Adoption Patterns and Governance of AI Coding Assistants in Practice

Wednesday 2 December

10:50, J1610 (industry): Tingberg, Sannicolo, Runeson, Aladjoff, Can TDD improve the quality of LLM-migrated code? A pilot study in an industrial context
13:30, J1610 (industry): Bexell, Gullstrand Heander, Söderberg, Eldh, Orchestrating AI-Assisted Code Remediation: Socio-Technical Bottlenecks in a Large Industrial Repository
13:50, J1610 (industry): Mäntysaari, Saltan, Siemon, From Generated Patches to Verified Fixes: A Validation-Aware Pipeline for LLM-Based Technical Debt Remediation (title truncated in the programme)
14:10, J1610 (research): Lepre, Leotta, Ricca, POMify: LLM-Assisted Refactoring of Selenium Test Suites toward the Page Object Model
14:25, J1620 (research): Xueping Gao, Evidence-Calibrated Runtime Reconstruction for Agent Skills Across Heterogeneous Coding Agents
14:10, J1620 (short): Nyberg, Mattsson, Borg, Britto, Saini, Machine Learning-Based Detection of MCP Attacks
15:50, J1610 (research): Cheng, Kim, Jiang, Zhang, Fu, Truong, Thi Huyen, Khomh, Yoshioka, Ubayashi, Washizaki, Does Agent Configuration Matter? An Empirical Study of Process-Level Quality Adaptation (title truncated in the programme)
16:30, J1610 (research): Villamizar, Fischbach, Şahin, Frattini, Ferrari, Korn, Vogelsang, Mendez, On the Impact of Requirement Smells in LLM-Based Code Generation

5. Where our research meets it

Harness first built a small shop model, oracle, invariants and fault injector first, then planted fifteen faults (seven mutants, two page faults, six TLA+ faults) and recorded which layer saw each; one shop, one lab, one day of runs.
When hardening comes due argues that hardening of agent-written code is due when the work moves to a new context (audience, lifetime or authority), with six triggers drawn from failures in experiments 016 and 017 of one lab repository.
From spreadsheet to service proposes two ladders (correctness and operations) and four controls for making small vibe-coded apps safe to depend on; its numbers come from one lab project, not a production system.
Elenctic vibe code review argues for reviewing AI-generated code by structured questioning and a hardening ladder rather than prohibition, illustrated on three agent-built apps.
an empirical venue would expect what these notes lack: more than one subject application and more than one lab, faults seeded by someone other than the harness author, statistical analysis of the outcomes, and an explicit threats-to-validity section.