SociaLLM Engineering: On Manipulating AI Agents and what we can do about it