There's a fundamental mismatch in how we approach coding agents versus enterprise agents. The reality is simple: code exists everywhere online in abundance, and unlike other systems, it's inherently verifiable and testable. This creates wildly different expectations. Enterprise agents operate under constraints—they need to justify decisions, show their work, prove reliability. But coding agents benefit from an opposite dynamic: infinite training data combined with measurable outcomes. You can actually validate whether the output works or fails. That gap between what people expect and what these systems actually deliver tends to be where all the friction happens.