There are a handful of markets that are large and liquid enough to fully move to electronic trading, like spot FX, vanilla interest rate swaps, perhaps treasuries trading.
But most other markets are very much relationship driven. For instance a trader will make a market on illiquid bonds based on what he thinks is the appetite from the short list of potential buyers, and that's based on sales people and brokers discussing with these investors and feeding him with feedback.
Very similar for mergers and acquisition which is mentioned in the article. It's probably something like 1/3 relationship and advisory to the client, 1/3 going through legal and regulatory issues, 1/3 working out financials, which are often based on unique and complex accounting standards, market specific idiosyncrasies, etc. Good luck training an AI algorithm to do that. And with what data?
There is massive room for improvement in the financial industry. But most likely in the more boring areas of operations and settlement (cash settlements for instance are massively labor intensive as there are always incidents, system issues, accounts details changing, etc).
My personal experience of corporate IT in a large international bank is that there is a handful of good people but an enormous army of average to well below average people who do not know the difference between cash and capital other than it starts with a c, i.e. are completely uninterested by the domain they work on, and aren't good technically either. And these will ensure that outside of a few players (including probably goldman), we are guaranteed that automation is far far away no matter how much money we throw at it...