The upgrade of the VLTI infrastructure for the 2nd generation instruments is now complete with the transformation of the laboratory, and installation of star separators on both the 1.8-m Auxiliary Telescopes (ATs) and the 8-m Unit Telescopes (UTs). The Gravity fringe tracker has had a full semester of commissioning on the ATs, and a first look at the UTs. The CIAO infrared wavefront sensor is about to demonstrate its performance relative to the visible wavefront sensor MACAO. First astrometric measurements on the ATs and astrometric qualification of the UTs are on-going. Now is a good time to revisit the performance roadmap for VLTI that was initiated in 2014, which aimed at coherently driving the developments of the interferometer, and especially its performance, in support to the new generation of instruments: Gravity and MATISSE.

We present the latest update of the European Southern Observatory's Very Large Telescope interferometer (VLTI). The operations of VLTI have greatly improved in the past years: reduction of the execution time; better offering of telescopes configurations; improvements on AMBER limiting magnitudes; study of polarization effects and control for single mode fibres; fringe tracking real time data, etc. We present some of these improvements and also quantify the operational improvements using a performance metric. We take the opportunity of the first decade of operations to reflect on the VLTI community which is analyzed quantitatively and qualitatively. Finally, we present briefly the preparatory work for the arrival of the second generation instruments GRAVITY and MATISSE.

Since April 2011, realtime fringe tracking data are recorded simultaneously with data from the VLTI/AMBER interferometric beam combiner. Not only this offers possibilities to post-process AMBER reduced data to obtain more accurate interferometric quantities, it also allows to estimate the performance of the fringe tracking a function of the conditions of seeing, coherence time, flux, etc. First we propose to define fringe tracking performance metrics in the AMBER context, in particular as a function of AMBER's integration time. The main idea is to determine the optimal exposure time for AMBER: short exposures are dominated by readout noise and fringes in long exposures are completely smeared out. Then we present this performance metrics correlated with Paranal local ASM (Ambient Site Monitor) measurements, such as seeing, coherence time or wind speed for example. Finally, we also present some preliminary results of attempts to model and predict fringe tracking performances, using Artificial Neural Networks.