Note: Similar to Running Notes on RAG.
AI Agents will likely shine in areas where there is demand for solutions, but is constrained by demand on talent to do so. Good examples are coding agents, security agents, legal agents etc., In these areas, the demand for solving problems exceeds the talent available.
The playbook for building ai-agents is to deep on context engineering (bring all the data) required for the vertical or domain, figure out the User Experience that fits into the existing workflows naturally, and connect to relevant data sources and tools.
We are trending to more agents given narrowly scoped, well defined narrow tasks. These will be defined by professionals — martin_casado.
I think when combined with smol-llms, they can be unobtrusive doing their well defined tasks in the background, without much human intervention. “Manage by exception” or “manage by failure” instead of active human feedback. Human attention in precious, and should be overwhelmed with attention to even more computers.