Artificial intelligence is increasingly connected to automated synthesis, analytical instrumentation, experiment planning, and closed-loop scientific discovery. However, computational competence does not by itself establish the capacity to conduct pharmaceutical experiments. Laboratory work is embodied: every action is conditioned by equipment availability, material identity, container state, spatial reachability, procedural dependencies, timing requirements, contamination controls, hazard boundaries, measurement uncertainty, and human authority. Current autonomous-laboratory architectures demonstrate important elements of planning, robotic execution, sensing, and iterative optimization, yet these capabilities are commonly evaluated separately or summarized through narrow measures of task completion or optimization efficiency. This article develops an original design-principles construct, termed Constraint-Grounded Embodied Laboratory Intelligence, to organize the conditions under which artificial intelligence may progress from generating experimental proposals to participating in bounded pharmaceutical experimentation. The construct represents laboratory intelligence as a continuously updated relation among equipment affordances, material and sample states, executable procedures, temporal commitments, safety envelopes, epistemic provenance, perception, manipulation, adaptive execution, anomaly interpretation, and human takeover. The article argues that no single benchmark, planning score, manipulation-success measure, or model-confidence estimate can establish autonomous laboratory competence. Instead, evaluation must examine whether computational representations remain synchronized with physical laboratory conditions, whether proposed actions satisfy procedural and safety constraints, whether deviations are detected and interpreted appropriately, and whether humans can intervene before consequences become irreversible. The contribution is conceptual rather than empirically validated. Its applicability will depend on laboratory type, pharmaceutical modality, instrumentation, materials, local governance, and prospective evaluation. The proposed framework may nevertheless support more rigorous system design, reporting, comparison, and staged validation of embodied artificial intelligence for pharmaceutical research.