Elvin Aghammadzada describes the limits of loading extensive instructions and tool definitions into every agent session. He argues that irrelevant context can make reliable task execution harder.
Elvin Aghammadzada presents skillsAn AI agent skill is a reusable package of instructions, resources, and tool guidance for performing a bounded kind of work. as reusable instructions that an agent can load progressively. He compares that approach with MCP servers, which can provide access to remote tools, data and resources.
The talk also discusses risks of self-modifying and third-party skills, including prompt injection and limited verification. Elvin Aghammadzada recommends treating skills as software that needs testing, evaluationEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. and versioning.
Watch on YouTube
