From Physical Review of Letters: https://journals.aps.org/prl/abstract/10.1103/v6jq-m6sk

This is very interesting, because if this worked it would solve the greatest defect of the best atomic clocks.

The best optical atomic clocks provide only a correct average frequency. That means that if you have such an optical atomic clock you know with an extremely small uncertainty which is the frequency averaged over a long time interval, of days or weeks, but you know much less about the value of the frequency averaged over a short time, e.g. a second.

The instantaneous frequency and the frequency averaged over short time intervals are not determined by atomic properties, but they are determined by the length o f a resonant cavity. That length varies due to vibrations from the environment and due to temperature fluctuations. To minimize these length variations great care is taken for damping vibrations and for stabilizing the temperature, usually in a cryostat, but the accuracy required for an atomic clock is so great that even extremely small residual vibrations and temperature variations limit the performance of the atomic clock.

TFA describes a way to make a laser whose frequency is determined by the atoms whose stimulated emission is used, not by the resonant cavity where they are placed. A resonant cavity is always needed in a laser, to provide the positive feedback that is required for converting an amplifier by stimulated emission into an oscillator that provides a continuous output signal, without an input.

Because the frequency of such a laser is determined by atomic properties, e.g. of barium atoms in the example given in TFA, it is no longer necessary to have a system that tunes the resonance frequency of the laser cavity by measuring atomic resonances in some separate absorption cells, where such a tuning system can only ensure a correct value for the averaged laser frequency.