Cobus Greyling
Cobus Greyling is passionate about exploring the intersection of AI and language. From Language Models, AI Agents to Agentic Applications, Development Frameworks & Data-Centric Productivity Tools, I share insights and ideas on how these technologies are shaping the future.
Articles by this author
The scaling paradox of harness engineering
Harness engineering gives teams more control over agent performance. It also creates more dependencies that have to stay aligned for that performance to hold. At scale, keeping agent performance on track becomes very challenging manually as the system changes and production reveals new failure modes. Loop Engineering turns that into a continuous feedback process. Production performance informs what needs to change, and each change is validated against goals the agent is expected to deliver.
What is Hill Climbing?
Agentic AI changes software from executing predefined instructions to generating decisions and actions dynamically, making traditional debugging less effective. This blog explains hill climbing, an evaluation-driven approach to improving AI agents through observability, performance metrics, production traces, controlled experimentation, and continuous re-evaluation. It explores how teams can systematically improve probabilistic agent behavior while retaining deterministic controls, guardrails, and orchestration where they matter.
