KAIZEN | Scaling up the use of AI coding agents across software teams

Practical methods to maintain quality, security and control over software development

AI coding agents are becoming increasingly common in software development. However, moving from individual experiments to consistent practices across teams remains challenging. KAIZEN helps Flemish software companies adopt these tools in a structured and responsible way.

The project develops practical methods, guidance and demonstrators to integrate AI coding agents into everyday development workflows while maintaining software quality, security and long-term productivity. 
 

Target group

The project is primarily relevant for:

• Companies developing software products and introducing AI coding agents into their development processes  
• Software development teams looking to move beyond individual experiments and adopt AI coding agents more widely  
• Companies seeking to improve productivity while maintaining software quality and security  
• Technology providers developing AI coding tools and related services 

Context

AI coding agents such as Copilot, Cursor and Claude Code are becoming increasingly common in software development.Developers are increasingly experimenting with these tools and sometimes report significant short-term productivity gains. In practice, however, adoption is often ad hoc and uncoordinated. Companies must determine which tasks to delegate to AI agents, how to maintain software architecture, quality and security, and whether productivity gains are sustainable or simply shift effort from development to reviewing and validating code.

Scaling up the use of these agents across development teams brings additional challenges. AI coding agents need sufficient context, consistent quality rules and fast feedback loops to produce reliable results. Companies must also establish clear rules for security, data protection and intellectual property. 

Without established practices and reliable ways to measure impact, companies struggle to move beyond individual experiments and adopt AI coding agents across their teams. 
 

Objectives & results

KAIZEN helps Flemish software companies adopt AI coding agents in a structured and responsible way, from identifying relevant use cases to integrating these tools into everyday development workflows.

The project aims to:

• Help companies assess their readiness for AI coding agents and identify relevant use cases
• Develop practical methods for integrating AI agents into software development workflows
• Establish good practices for quality, security and human-agent collaboration
• Evaluate the impact of AI coding agents on development processes and productivity
• Support the organisational changes required for broader adoption

Key expected results

• An adoption framework and maturity model to assess readiness, identify relevant use cases and progress from experimentation to broader adoption  
• Practical playbooks for AI-assisted software development, covering the integration of AI agents into development environments and collaboration between developers and AI agents  
• Demonstrators for representative use cases, including specification-driven development and work with legacy codebases  
• Training and awareness resources, including exercises, case studies, videos, articles, workshops and a guide for software entrepreneurs 
   

Approach

KAIZEN combines applied research with direct input from software companies through three complementary workstreams.

1. Understanding adoption needs 
   Interviews with software companies and an analysis of current practices help identify existing approaches, barriers and promising use cases. These insights provide the basis for the adoption framework and maturity model.
2. Developing practical guidance 
   The project investigates key aspects of AI-assisted development, including quality, security, tool integration, feedback loops and human-agent collaboration. The findings are used to develop practical playbooks.
3. Testing in realistic software environments 
   The partners develop demonstrators for representative use cases selected together with the project advisory group. These demonstrators help validate the proposed approaches and identify lessons learned and good practices. 
    

Funding

• Funding agency: VLAIO
• Project type: COOCK+ project