FIELD NOTES / 05 CHAPTERSAI · SYSTEMS · HUMAN REVIEW
01PROMPT LIBRARY
Starting with prompts
ChatGPT was my entry point. I wrote and refined offers and service descriptions, summarised text and built a prompt library. Variable prompts and explicit prompt chains became ways to turn those individual steps into repeatable workflows.
02SEPARATE CRITERIA
Learning the limits
As an AI Competence Center Lead, I explored reviewing data-processing agreements under Article 28 GDPR. Reviewing whole contracts against every criterion proved unreliable, so I moved to criterion-by-criterion prompt chains. Policy generation combined standard-based requirements, including ISO 27001, with the user’s choices about their organisation and the aspects to cover.
03CONNECTED SOURCES
Working with knowledge
Source-backed questionnaire answers led me to knowledge retrieval. For dissertation work, I parsed large document collections. Source quality and traceability became central concerns, alongside how search, structured queries and direct access to files could work together.
04AGENTIC SYSTEMS
Building with agents
I first worked with a developer team on AI tools. Claude Code then let me build my own experiments around Agentic OS, including a working CRM prototype and voice control. Building a voice agent moved the interaction beyond a graphical dashboard.
05HUMAN REVIEW
Examining the results
Starting parallel tasks by voice made processing, prioritising and combining their results a new bottleneck. That experience brought human review into focus. My independent AI Security research explores how systems behave, where they fail and how their outputs can be examined.