This paper details the progressive steps taken in enhancing the onboard automation of ESA's Gaia mission in response to in-flight experience. Tasked with mapping 1 billion stars to unprecedented precision (to the micro-arc-second level, comparable to the width of a smart phone on the Moon as viewed from Earth). ESA's Science cornerstone mission is expected to also discover and chart 100,000's of new objects including near Earth asteroids, exoplanets, brown dwarfs and quasars. After a flawless launch 19 Dec 2013, Gaia was brought the circa 1.5 million kms into L2 via a sequence of orbit transfer manoeuvres. Starting in parallel to this, and lasting 6 months, the full spacecraft was commissioned and brought gradually up to full performance. Since Q3 2014, Gaia has been in the routine operations phase, gathering the enormous Astronomical data set for the multiple map releases planned throughout the mission lifetime and beyond. During commissioning, and later routine operations, ground responded to a number of challenges with the aim of efficiently maximizing mission performance. Measures were taken to increase onboard autonomy in a number of areas. These have been of two main types; one being efficient increase of mission data return (e.g. due to onboard performance in some areas being higher than expected), and the second responding to repeatable anomalies (i.e. autonomously recovering and thereby limiting their impact). Such repeatable anomalies have no permanent impact on the performance of the mission and can have their origin in various subsystems in the space and/or ground segment. This paper details various examples of onboard autonomy enhancements that have been implemented since launch. Their design, validation and implementation are described, along with an assessment of some of the onboard PUS services used (e.g. OBCPs, event-action, TC sequences) with respect to their usefulness to ground operations teams. The trade-offs and logic used when deciding firstly what to automate (and what to leave to manual operations), and also where to automate (onboard or on ground using the closed loop MATIS system), is also presented. Specific examples in the paper will include: dealing with a significant increase of data volume (circa 45%) to be downlinked to ground; automatically coping with local bad weather events at the ground stations; dealing with repeatable onboard anomalies ranging from the relatively benign (those temporarily impacting the precise thermal balance) to the more severe (those that can trigger Safe Mode and produce data losses in the range of weeks).