From a useful application to an agent that can be evaluated.
I develop and evaluate AI agents in practice. My approach connects useful applications, required skills, data discovery, infrastructure and data preparation. It draws on 250 formal, documented interviews on introducing AI in companies.
Practice as the starting point
250 formal, documented interviews on approaches to introducing AI in companies inform my work (October 2026). I combine qualitative analysis and literature with building and testing agents. These interviews are distinct from the 400 professional conversations informing my transformation work.
From use case to evaluation
I developed a practical approach connecting use-case selection and required skills with data discovery, infrastructure, data cleaning and restructuring. Qualitative interviews inform requirements; quantitative testing examines agent reliability. The central question is how to evaluate agents meaningfully so that evidence supports product decisions.
Evaluation for product decisions
I define what a good answer or action looks like, test typical failure cases and compare results with the value delivered in a real workflow. Qualitative requirements and quantitative testing must inform the same product decisions.
Further research
My dissertation proposal examines responsible organisational adoption of autonomous agents: design criteria, governance, auditability, adoption approaches and target operating models, including skills development and changing responsibilities. It describes planned research alongside ongoing hands-on product development.