Agent Evolution: How agents are advancing from prompts to self-improving loops
About this session
AI world is evolving fast, from prompt engineering to context engineering, then harness engineering to loop engineering. This also shows a clear advancement in industry use cases.
The next generation of agents will not be improved by better prompts alone. They will improve through harness evolution: better tools, skills, memory, compaction, artifacts, evals, observability, and feedback loops.
In this talk, I’ll show why many production agents plateau even as models improve. Their prompts, tool lists, evals, and orchestration were often built for older model behavior. A naive model swap can look underwhelming because the harness is hiding the improvement.
Speaker
Key takeaways
- Learn how modern AI Agent Architectures
- Learn how to evolve your agent from old-fashioned architectures to modern primitives
- Harness Engineering