Mesoscale architectures/microstructures for enhancing mechanical performance are already employed in steels. In next-generation high-strength/high-toughness steels, the austenite/martensite (fcc/bcc) interface is the dominant microstructural feature controlling properties, but the structure and motion of this interface remain uncertain. Here, an atomistic fcc-bcc iron interface is constructed that completely matches experimental observations. The interface reveals a defect structure differing from longstanding assumptions and violating conditions believed essential for a glissile interface. Based on the new interface structure, we revise (i) the conditions for achieving a glissile interface and (ii) the double-shear theory of lath martensites. The new parameter-free theory is in near-perfect agreement with simulations and experiments, and thus allows for guided design of materials with higher toughness. The new structure also has implications for alloying and hydrogen embrittlement. The adroit application of atomistic modeling tools thus expands the possibilities for design and control of mesoscale architecture in high-performance steels.