Dexterous manipulation - especially when objects are small, cluttered, or partially occluded - remains a complex challenge in robotics. In this talk, I present three approaches to tackling this problem using DenseTact, a family of soft, vision‑based tactile sensors. First, I show how end‑effector design, particularly the fingertip, can provide rich information during contact with the environment. The design of DenseTact and its successive improvements demonstrate how to sense and enhance manipulation through high‑resolution measurements of surface deformation and force. Next, I describe a physics‑aware calibration pipeline for the optical tactile sensor that reconstructs the full stress field across the gel information crucial for manipulating diverse objects and environments. With these components in place, I show how the calibrated sensors broaden the range of manipulation behaviors. Integrated into various types of grippers, such as the Leap Hand and an 8‑DOF rolling gripper, DenseTact’s tactile information improve the grasping of small objects and enable complex, contact‑rich behaviors via behavior cloning.