Why Identity Matters More Than Instructions
Instructions handle the expected. Identity handles the unexpected.
If you write an instruction for every possible scenario, you get a long, brittle prompt that breaks the moment reality deviates from your script. And reality always deviates. The customer asks something you didn't anticipate. The data comes in a format you didn't plan for. The previous agent in the pipeline produces output that doesn't match expectations. In those moments, an agent with instructions and no identity does one of two things: it freezes (asks for clarification when none is available) or it defaults to generic behavior (produces something safe, bland, and unhelpful).
An agent with a well-designed identity does something different. It reasons from its identity. "I am the kind of agent that values precision over speed. When I encounter ambiguous input, my instinct is to flag the ambiguity rather than guess. My primary obligation is to protect downstream agents from acting on uncertain data." That agent doesn't need an instruction for this specific scenario. Its identity generates the appropriate response.
Think of it this way. You've worked with people who are good at their jobs. When something unexpected happens, they don't consult the employee handbook. They make a judgment call based on who they are, what they value, and how they think. The best judgment calls come from people with strong professional identities, people who know what they stand for and what they won't compromise on. Agent identity works the same way.
This is not anthropomorphism. You are not giving your agent a "personality" for entertainment. You are giving it a cognitive architecture that produces consistent, predictable behavior across novel situations. The identity is functional, not decorative.