Strong suspicion that engineering work on agents is mainly a short-term tactic that we'll need to use to launch quickly and learn; eventually we'll discard much of that code and hopefully benefit from the market presence and insight gained while others waited.