Active · Grand Solutions · Innovation Fund Denmark

Research / AI in Software Engineering / Funded Project

AI4SE1DK: Human-Centred AI for Software Engineering

A national four-year research programme positioning Denmark as a global reference for trustworthy, human-centred, and sustainable adoption of AI in software engineering. The programme closes the current adoption gap while strengthening the competitiveness of Danish software organisations.

FunderInnovation Fund Denmark
Budget30.4 MDKK total
DurationApr 2025 – Mar 2029
Partners14 organisations
≈20M
DKK from Innovation Fund
14
Consortium partners
10K+
Practitioners targeted
48
Months / 8 work packages

Overview

AI4SE1DK is a Grand Solutions project funded by Innovation Fund Denmark, running from 1 April 2025 to 31 March 2029. The project addresses a structural gap in the Danish software sector. Although the sector comprises more than 100,000 professionals and generates roughly 313 billion DKK in annual turnover, only about half of Danish software organisations currently consider using AI for software development. The programme aims to close that gap in a manner consistent with Scandinavian values of transparency, worker wellbeing, ethical conduct, and environmental sustainability.

The initiative brings together six universities (SDU, Aalborg University, University of Copenhagen, Aarhus University, IT University of Copenhagen, and DTU), one GTS institute (Alexandra Institute), and seven industry and public-sector organisations (Umbraco, SIG, DCR Solutions, Ericsson, Unik System Design, Systematic, and ESS). Beyond the core consortium, an implementation board comprising IDA, Prosa, IT-Branchen, DI Digital, DigitalLead, DIREC, IT-Vest, and UCL is responsible for scaling outcomes across the wider Danish software sector.

Wide adoption of AI for software engineering is estimated to unlock approximately 70 billion DKK in value for Denmark, through productivity gains, new products and services, and reduced time-to-market. AI4SE1DK provides the scientific and practical foundations to realise that potential responsibly.

Research Objectives

The programme is structured around three SMART objectives spanning adoption, technical development, and national-scale dissemination. Each objective is paired with measurable targets validated against established frameworks.

O1

Close the Adoption Gap

Advance AI4SE practices to readiness level SRL7 through structured collaboration between universities, research and technology organisations, and Danish software companies. The objective explicitly links AI adoption to workforce wellbeing rather than pure automation, using the SPACE framework and the Team Climate Inventory as measurement instruments.

SRL 7Target readiness
+15%Team climate
500Developer survey
O2

Enhance Technical Adoption and Trustworthiness

Develop AI4SE methods and tools tailored to Danish organisations, reaching Technology Readiness Level 6 by project end. Technical outputs include conversational agents, company-specific code assistants, and scalable trustworthiness assessment methods combining static and dynamic analysis, formal methods, and AI-based analysis.

TRL 6Technology readiness
+30%Productivity target
+10%AI code adoption
+15%Trustworthiness scale
O3

National Outreach and Implementation

Through the implementation board, deliver concrete practices, tools, and guidance to at least ten percent of the Danish software industry, approximately 470 to 477 organisations and more than 10,000 practitioners. Project outcomes are integrated into educational curricula for roughly 6,500 annual IT enrolments, professional networks, and SME-oriented activities to ensure diffusion beyond the core consortium.

10%Industry reach
470+Organisations
6,500Annual IT enrolments

Unmet Needs

The programme identifies four tightly connected structural challenges in Danish AI4SE adoption. These are not independent problems. The knowledge gap reinforces technical hesitancy, which compounds trust deficits, all within a context where ethical and sustainability concerns remain insufficiently addressed.

C1

Knowledge Gap

Limited context-sensitive guidance on integrating AI into existing software engineering workflows leads to poor or hesitant adoption, even where motivation is present.

C2

Technical Adaptability

Generic AI tools such as code assistants are difficult to tailor to company-specific architectures, internal quality standards, and established development processes.

C3

Trust and Verification

Widespread scepticism about the reliability, maintainability, and regulatory compliance of AI-generated artefacts, including the EU AI Act, combined with a lack of scalable oversight mechanisms.

C4

Ethical and Sustainable AI

Little systematic attention to the job impact and carbon footprint of AI4SE, despite Denmark’s strong institutional commitment to ethics and environmental sustainability.

Industrial Use Cases

The four challenges are operationalised through industrial use cases that span the full software lifecycle. Each use case is embedded within a partner company, enabling co-creation, longitudinal action research, and direct validation of project outputs in production settings.

UC1

AI-Native Development Processes with Autonomous Agents

Investigates the integration of autonomous AI agents into agile software development processes at scale, including AI-driven project management and scientific software generation.

Partners: Ericsson · ESS
UC2

Tailored AI4SE for Product and Quality Standards

Develops company-specific AI tools incorporating internal codebases, domain-specific languages, and quality standards to improve code generation and review for established software products.

Partners: Umbraco · Unik System Design
UC3

AI-Supported No-Code Development for Regulated Domains

Builds conversational AI agents enabling process modelling and application development without traditional coding, targeting regulated healthcare and compliance contexts.

Partners: DCR Solutions
UC4

Trustworthy AI in High-Criticality Contexts

Advances scalable assessment and verification of AI-generated software in safety-critical systems, including “assessment as a service” and integration of AI agents into CMMI Level 5 development processes.

Partners: Systematic · SIG (Software Improvement Group)

Methodology and Work Packages

AI4SE1DK applies a combination of empirical software engineering, constructive research, and longitudinal action research. The methodology reflects a commitment to producing both scientific knowledge (papers, datasets, models) and practical artefacts (tools, guidelines, proof-of-concept deployments) in tandem. Eight work packages link foundational studies, technology development, use-case validation, and national dissemination into a coherent programme.

WP1

Project Management

Governance, coordination, steering committee, and quality assurance across the programme.

WP2

Empirical Foundations

Cross-sectional survey (n=500) and longitudinal action research mapping AI4SE practices, needs, and sustainability.

WP3

Tailored AI Tools

Conversational agents and specialised AI tools incorporating company codebases, DSLs, and internal standards.

WP4

AI Software Assessment

Scalable, partly automated assessment combining static and dynamic analysis, formal methods, and AI-based techniques.

WP5

AI-Native Processes (UC1)

Generative AI agents for agile project management and scientific software generation at Ericsson and ESS.

WP6

Company-Specific Tooling (UC2–3)

Domain-specific code generation and no-code conversational agents validated at Umbraco, Unik, and DCR.

WP7

Trustworthy AI (UC4)

Assessment as a service and AI agents in CMMI5 processes, validated at Systematic and SIG.

WP8

National Adoption

Dissemination, proof-of-concept projects with at least six external companies, and co-organised industry events.

Consortium and Partners

The consortium spans six universities, one GTS institute, and seven companies and public-sector organisations. This composition ensures that fundamental research, applied development, and direct industrial validation proceed in parallel rather than sequentially.