U-Define: Designing User Workflows for Hard and Soft Constraints in LLM-Based Planning

U-Define's core insight is that users don't think in numeric weights. The research shows that binary constraint categories (hard vs. soft) paired with transparent verification logic let non-technical stakeholders encode real-world requirements without learning formal constraint syntax, a usability finding that prior work on constraint systems largely overlooked.

This connects directly to the pattern established by RunAgent (May 1) and the procedural execution diagnostic study (May 1). Those papers identified execution reliability and step-fidelity as the actual bottlenecks in LLM deployment, not model capability. U-Define extends that insight upstream: before execution can be reliable, users must be able to specify intent clearly. The chart validation pipeline (May 1) also decomposed a complex task into inspectable stages with explicit gates. U-Define applies the same principle to the constraint-specification layer itself, treating user intent encoding as a design problem rather than assuming users will adapt to rigid frameworks.

If enterprises adopting U-Define report measurable reductions in plan rejection rates or rework cycles compared to numeric-weight baselines within 6 months, that validates the usability claim. If instead adoption stalls because users still struggle with the binary model, the research remains academically interesting but signals that constraint specification remains unsolved.