Fred Ramberg on AI, Agentic Trading and the Necessary Past

Jun 23
11 min read

June 24, 2026 | In this issue of The Institutional Risk Analyst, we catch up with an old friend and colleague, Fred Ramberg, a serial entrepreneur who we had the pleasure of working with in the 2000s in the world of semiconductors and specialty capital equipment. Fred holds multiple patents and worked in the world of machine learning long before the deal-maker machinery behind the Wall Street narrative coined the term “AI.” Fred is the sort of sophisticated technologist who writes his own patent descriptions and understands the different physical and virtual layers of technology in a profound way.

The IRA: Fred, great to reconnect. Did we see that you somehow found time to start a new coffee company while you remain focused on what we now call AI?? We recall your fondness for strong coffee.

Ramberg: Keurig's base patents had expired. And we said, why do you put coffee in this little basket? Why not put it in the lid? So we have a lid, standard size, standard cup. You put it under your Keurig machine or a very simple machine that we buy from China for six dollars and ninety-five cents. And it makes coffee.

The IRA: Great idea. You know what? You could push this into the hospitality channel with Marriott (MAR). They use many different funny little machines that, you know, produce hot water and you can make tea with them too.

Ramberg: My board is comprised of two early employees of Keurig. It was very interesting how they started. The first market for Keurig was not mass market humans, direct to consumer. Their first market was small and mid-sized business. They would go in and say, “put this in your break room.” Then they fulfilled through W.B. Mason and then ultimately Office Depot and Staples. They only had to sell to just three customers. They shipped out coffee pods to three customers which fulfilled the orders.

The IRA: Sounds like a great story. W.B. Mason is a fascinating family-owned business that carries everything for the office. They sell a lot of coffee. And coffee is a vast global market.

Ramberg: It's kind of a big one. The reason we stuck with it and we've now filed two patents, BTW, is that Keurig ultimately sold to Dr. Pepper for eighteen billion dollars. Coffee is not a small market. And Keurig is not the number one coffee in the world, believe it or not. It's JDE Pete's NV (JDEPY).

The IRA: So besides coffee, we take it that you are still focused on technology and semis?

Ramberg: I have three recent patent applications in AI, but I've got seven issued. They're assigned but they're issued. Bottomline purchased our machine learning company in 2014. When Bottomline later sold to private equity, they virtually abandoned the patent portfolio, including my patents. And it's one of the most valuable AI patent portfolios I've ever seen. But we were paid nicely.

The IRA: That's a striking comment. Why do you believe that Bottomline abandoned the patents?

Ramberg: Well, because they sold themselves as a payments company. We did banking software and payments. And in 2014, the patents were seen as non-core. Right. I tried to license one of the patents back, but they wouldn't even enter a discussion.

The IRA: Yeah, it's funny how people's behavior changes when you're trying to sell yourself to a big corporate takeout. We’ve seen this over and over again with entrepreneurs. We know a fascinating AI play in residential mortgage due- diligence that is going through this process now.

Ramberg: Wind the clock forward a decade and today it's very hard for startups to get into the technology channel.. It's hard to get into business with the company that we sold back then. We sold our machine learning predictive analytics company more than a decade ago. We were fortunate to have a financial backer, Mark Rosenblatt.

The IRA: Last week you started talking to me about the AI market and stocks and all the rest of it. I had to actually go online and do a little reading about agentic agents. We find “agentic trading” quite amusing because of the obvious cinematic metaphors. Call me Neo. And we are friends with Michael Green, Chief Strategist at Simplify Asset Management. He was one of the people who years back started teaching us about passive investing via exchange traded funds (ETFs). The steady bid from passive strategies tends to push stock prices up – until is doesn’t, like yesterday (06/23/26) in AI names. How does agentic trading impact stocks? Is agentic trading a large part of the AI surge due to this sort of trading? Is it significant?

Ramberg: I don't know how significant agentic trading is today, but it is big. Trading algorithmically represents, in my reading, 60% plus of daily volumes, trades which are algorithmically driven. But algorithmic trading has been around for a long time. The bots trade with a purpose. The algorithms recommend, you know, move you with a purpose. And they don't work with one another. However, as they become better with machine learning, and I won't use the term “AI” quite yet, they tend to correlate.

The IRA: That sounds like more passive bid for stocks, but this correlation goes both ways, correct?

Ramberg: What the algorithm senses for data is the same data as the other algorithms are seeing. So they tend to recommend the same actions based on that same data. What that does is amplify moves. So should you be worried about amplification of moves where what might have been a one or two percent move is now a five or six percent move. They're all trading on the same data in the same direction at the same time.

The IRA: Insider trading by machines is still insider trading. This discussion of AI and agents reminds us of a question we wanted to ask. Do we need more than one large language model (LLM), Fred? Are the developers of LLMs actually creating value by staring at the same dataset?

Ramberg: Yes.

The IRA: So each LLM is different even though they use the same data? We are suing Anthropic, BTW, for stealing several books. Wonderful people.

Ramberg: Yes. Absolutely. They all build an ontology in a very different way. Do they get to the same place often? Yeah. But do we need more than one? Yeah, definitely.

The IRA: So in addition to passive strategies, we now have to worry about free agentic agents driven by a multiplicity of AI models?

Ramberg: Yes. That is known and it's an interesting question. What's not well studied, though, but is talked about, for example, by the the Bank for International Settlement, is what happens when agents trade with agents without any intervention or any input from human beings.

The IRA: Indeed. So these little AI bots can also come together with no human intervention. In other words, they can come to the same conclusion and then act together even though they're not really acting in concert. Isn't that illegal?

Ramberg: Yes they do. They correlate a lot because they're operating on the same model, with the same data at the same time. That is what's called accidental concert. So, in accidental concert, yes, the agents work together and that causes the amplification I was talking about. If everybody does the same thing at the same time, um, moves can be larger.

The IRA: So what should people be worried about?

Ramberg: Anybody focused on investing not concerned about agentic trading is is fooling themselves. And it comes from a discussion I had with one of my sons about what may be happening inside virtual chat rooms that now exist that humans cannot enter. Only AI agents can enter these chat rooms. And they work with each other, not necessarily on trading and on strategies, but they work with each other on numbers and issues. Humans are barred from these chat rooms. So what I started to look for was rather than just amplification, moves that were, large on a daily scale in both directions. So now, if you go back to the days of pump and dump on Wall Street, what you might have going on, if you can detect it, is AI agents working with one another to pump up the price of an issue and then at some point short it and sell it down. Or short it to start off with and then buy it up.

The IRA: For example?

Ramberg: If you bought Micron Technology (MU) last Thursday at the close, sold it by the open Tuesday and shorted it, and then shorted it some more - by the open Wednesday you would be up 30%. In one trading session! Q.E.D. And this type of behavior is increasingly likely in AI and AI adjacent stocks where the moves are unreasonable. The fundamentals of the company would not suggest it. You and I talked about MU last week and covered the stock for years. The fundamentals of Micron as a company would not suggest sometimes 10% daily swings between the short and then the close.

The IRA: Indeed. We had several sleeping semi positions explode the the past year. The markets are taking profits in AI today in a way that suggests that the passive strategies are adding to the volatility. But you believe that agentic trading may be responsible for some of the short-term volatility we’ve seen?

Ramberg: Yes. And that's where I started to examine whether, you know, it looked like agents were acting in concert, I'll use the word collusion, that's a legal word, with one another. And though nobody's proved collusion, it's in the research for agents. There are no patents or IP filings that are published that reference it. However, there is significant academic research to see what the implications of collusion among agents would be.

The IRA: Well, this topic of agents really got our interest after our discussion last week. We started wondering about banks because there are times when these obscure small cap entities just fly for no apparent reason. Last year, for example, Lending Club (LC), which is one of the smallest banks in the top 100, took off. It was the best performing banks for six months. And it's not that it has great fundamentals. It's almost too small for most people to own and yet off it went like a rocket. Another one is SoFi Technologies (SOFI), which was the leading bank stock for most of last year. And now it's at the bottom of the group, by the way. You see this behavior and you think to yourself, who is doing this?

Ramberg: I've recognized a meme in certain stocks that you just described. And the question then becomes, who's doing it? And the answer might be nobody. The answer might be that agents have been let loose. They have identified factors based on parameters they've been given, volume, price, you know, and any other metric you might find. The agents have found data where they say, “this is likely to be susceptible to doing this.” And then they work together to do it. And that's all happening in this room that humans are not allowed to go into.

The IRA: Why do we feel like we’re in a sequel to the film "The Matrix?" The owners of the servers that support agentic trading are manipulating markets, plain and simple. Yet these agents act only based on rules.

Ramberg: Yes, but the agents don't even necessarily have to talk to each other. You know, the models are such that they probably both got to the same stock at the same time. So yeah, this is based on everything we can do, not that has been done, based on everything we can do, it's likely that this is a good candidate. And then they do it. Because their command isn't to go manipulate XYZ Corp, the command is, go find an issue that allows us to make a large amount of money.

The IRA: How does this world of AI and independent agents evolve?

Ramberg: So, what's coming, and I'll keep this short because you've probably heard me say it. But this looks like spreadsheets in the 1990s. There were seven makers of spreadsheets, and it was all a technology fight. This one made a mistake in pi in the thirty-second decimal place, and you know, it was about whose product is better than whose. And that continued forward. But, how many spreadsheets are there today?

The IRA: One at Microsoft (MSFT). Two if you count Alphabet's Google (GOOG).

Ramberg: Correct. But someday AI is going to be an integrated solution like Microsoft Office. And so as we move towards that end, the hype around AI is going to continue for some time to pull the adjacents, the MUs of the world.

The IRA: We think it is tough to pick winners and losers at this stage. Your comments about Oracle (ORCL) and Adobe (NMS) resonated. But will AI continue to be about merely understanding the past or will it start to focus on the future and how to create a specific desired future?

Ramberg: Bingo. I did some work on this aspect of AI and I wrote a patent that said, I need this to happen. What has to happen first? And we coined a term called a "necessary past." And so what we did is to articulate what we could do to create a necessary past. And lo and behold, the result pops out. So that's closer to utilizing the type of technology that makes up AI than asking it what'll happen if this customer buys sneakers and laces and socks, what'll they buy next?

The IRA: So like the Avengers, we go back in time, collect the infinity stones and get a desired future? That sounds a little scary. And all done by independent agents?

Ramberg: Yes. Most if not all of the AI models are infringing on that patent we discussed earlier. And that's why I wanted to license it back.

The IRA: It is human nature to speculate, but not necessarily take a structured approach to the speculation. How does the structured world of AI help to give us specific outcomes? Is this even legal?

Ramberg: Right. The agents are smart enough to parse a necessary past that leads to a desired outcome. This is comparable to the boiler rooms in the 1930s that Joe Kennedy had to legislate against, well, regulate against as the first SEC chairman, right? Your father was the first biographer of Joe Kennedy. Well, that's what you see here. And they better get some strong regulation and strong detection for this. Because once they start asking the agent the question the right way, I want this to happen. This is what I'm going to make happen. What do I have to do? What has to happen for that to be the result? As soon as agents are let loose to answer that question, then you got a problem.

The IRA: Thanks Fred.

Fred Ramberg on AI, Agentic Trading and the Necessary Past

Recent Posts

Recent Posts

Comments

Subscribe to The Institutional Risk Analyst