The building construction industry accounts for a substantial share of global carbon emissions, and concrete is especially critical because it requires large amounts of cement. In mid-rise residential buildings, floor slabs typically represent around half of the concrete volume [1], highlighting the importance of reducing material in slabs to decrease embodied carbon in buildings. Component-level shape optimization of beams has demonstrated material savings of up to 50% [2], and recent work extends this approach to optimize beam shapes across an entire slab-beam layout [3]. However, current workflows rely on non-gradient-based optimization and finite-difference sensitivities, causing runtime to scale with the number of beams and limiting early-stage design exploration. This research aims to accelerate slab-beam shape optimization by reformulating the problem to enable gradient-based optimization with automatic differentiation, combining a differentiable geometric parametrization with differentiable reinforced-concrete constraints. Geometrically, the beam shape is parameterized by a spline along the span with a T-shaped section integrated into the slab to ensure structural connectivity. Spline control points define the design variables and enable a differentiable evaluation of mass and section properties [4]. In terms of concrete mechanics, standard strength and serviceability checks are reformulated to avoid non-smooth operators and discrete logic, like the governing min/max envelopes and max-deflection criteria. Capacity checks based on strain-compatibility equations are formulated as implicit relations, with gradients obtained via implicit differentiation. Runtime and solution quality are evaluated against a finite-difference baseline and demonstrated within a slab-beam layout optimization workflow. The resulting improvements support a broader early-stage exploration of design alternatives and provide a basis for surrogate modeling that can scale the approach to building-level optimization.