AI and the business of Money

 Long, long ago, at PayPal, in one of the conversations with the Chief Architect of PayPal, I did ask about our API strategy for partners. Most of our APIs were Open APIs. Anyone can integrate it easily. I was asking and making a case for making the access more accessible.

He went on to compare the Uber APIs vs us and said, you end up in a wrong place, it is bad still manageable. People know your address, a privacy concern still relatively manageable. Now, if you lose money, it is a different ball game. I still felt, we should make it easy to people to access. The leaders prided on being a risk company, meaning the strength of PayPal is about managing and protecting risk than a payments. 

Point is, when it comes to money, one can never be too careful. 

So, am surprised when I see people are surprised to see Goldman blocking access to certain generative AI LLM tools.

Let me clarify. I am not anti AI. If anything, I use AI left right and center. In fact, I believe this post in its final form is way better cause of AI. 

Let me also give an example. As I said, I use AI for a lot of things. Sometime back, I asked to build a portfolio with some of my criteria, thesis and asks.  It gave me a list of 23 stocks in April 2026. Fast forward to November2026. Their performance have ranged from +111% to -18.6% the bottom 8 losers at (-7.1% to -18.6% and winners ranging from +5.4% to +111% while nifty50 has been at +4.98% and Nifty500 at +6.45% and nifty 100 small cap at +11.2% if i had invested at equal weighted I would be at +21.09% [Yes, I did ask the AI to help me with the data and simulation]

However, when I asked the same AI yesterday to help me with a portfolio and a split up, it recommended me to buy 10 stocks at price of 455. When I went to execute, i saw the price was at 4550. Now, I had limited funds, I was watchful. Now, imagine, you had used AI, tasted success, it is right most of time, depending on what you asked, how you asked. (The operating key word is how) and you had set up an agentic AI to execute it. Probably there was human in the loop. (Or may be not)

If you had seen things that happen in capital market operations, despite the 4 eyes check, mistakes happen. With AI, people would tend to believe, machine may not make mistakes. 

Lo and behold, if a broker, trader had been executing the above using Ai without over sight (point is even if there is human oversight, human would tend to not exercise oversight)

Now, imagine the scale and even if it is a small error, and even the percentage of times it fails is also small, still the damage it could cause be huge.