This paper focuses on adaptive output-tracking flight control design problem with on-line extremum seeking command generation. The problem is motivated by the need to design an autopilot for autonomous close-coupled formation flight of two aircraft. Control design task is associated with the trailing aircraft only. Flying in formation, the trailing aircraft must constantly seek an optimal relative to the leader position that minimizes aerodynamic drag force induced by the wing tip vortices of the lead aircraft. The trailing aircraft dynamics are represented by two interconnected uncertain subsystems: 1) longitudinal (relative position, true airspeed) dynamics and 2) lateraldirectional (lateral position, bank angle, roll rate) dynamics. The two subsystems approximate the trailing aircraft motion in a closed-coupled formation with the leader. While the control command for the first subsystem is predetermined (desired longitudinal separation), the command for the second subsystem (desired lateral separation) is computed on-line such that the influence of the second subsystem on the first one is minimized. Using feedforward neural networks, direct adaptive model reference control, and on-line extremum seeking command generation, the proposed formation flight autopilot provides the trailing aircraft with bounded output tracking and minimizes effects of vortex uncertainty on the aircraft aerodynamic drag force. Boundedness of the tracking error signals in the closed-loop system is shown using Lyapunov direct method.