LLM Tool Failures: Only 3 Root Causes – Value, Condition, Intent

This piece argues that LLM tool failures stem from just three root causes: wrong values, missing conditions, and missed intent. It proposes a preflight checklist that separates what to ask from how to answer, using an external list to determine what the model asks about. By treating unknowns as slots to be filled via lookup rather than inference, and by making the goal the list of unknowns rather than the execution decision, the approach aims to eliminate invented values and ensure that missing information is surfaced to the user.