The recent DESI baryon acoustic oscillation (BAO) measurements suggest the presence of dynamical dark energy, as indicated by the $(w_0, w_a)$ parametrization of the equation of state. This implies a crossing of the phantom divide, which has significant theoretical implications due to the instability of quintessence models. The authors perform a non-parametric reconstruction of the equation of state and confirm the need for a crossing of the phantom divide. They explore the theory space of Horndeski gravity using an effective field theory (EFT) approach and find that non-minimal coupling to gravity is crucial for a stable phantom crossing. They propose the *Thawing Gravity* model, which has the same number of parameters as $w_0 w_a$CDM and naturally realizes non-minimal coupling. This model improves the fit over ΛCDM for DESI BAO, CMB, and type Ia Supernovae data. The *Thawing Gravity* model is a non-minimally coupled gravity theory that modifies the growth of perturbations but fits the joint DESI+CMB+SNIa dataset well.The recent DESI baryon acoustic oscillation (BAO) measurements suggest the presence of dynamical dark energy, as indicated by the $(w_0, w_a)$ parametrization of the equation of state. This implies a crossing of the phantom divide, which has significant theoretical implications due to the instability of quintessence models. The authors perform a non-parametric reconstruction of the equation of state and confirm the need for a crossing of the phantom divide. They explore the theory space of Horndeski gravity using an effective field theory (EFT) approach and find that non-minimal coupling to gravity is crucial for a stable phantom crossing. They propose the *Thawing Gravity* model, which has the same number of parameters as $w_0 w_a$CDM and naturally realizes non-minimal coupling. This model improves the fit over ΛCDM for DESI BAO, CMB, and type Ia Supernovae data. The *Thawing Gravity* model is a non-minimally coupled gravity theory that modifies the growth of perturbations but fits the joint DESI+CMB+SNIa dataset well.