Rewiring the
Investment Process

All right, we are about ready to get started. Thank you all so much for joining us and for joining the conference generally.

As you have undoubtedly heard, AI has been a major topic for our industry and markets since ChatGPT arrived in late 2022. Two years ago, at this same conference, we held a panel about investing in AI across the ecosystem: which companies and sectors looked compelling, and which were most at risk.

Today, we are revisiting the topic with a slightly different lens. Instead of asking where to invest, we want to look at how AI is changing the way an investment firm actually works. How do managers build the capability, find ideas, and process information using AI?

Joining me today are two managers at the forefront of building AI into their investment processes and their firms more broadly. I am pleased to welcome Divya Nettimi, Founder and CIO of Avala Global, and Soren Gandrud, CIO and Managing Partner of Jones Hill Capital.

Before we get started, I want to note that we will leave about 10 minutes at the end for Q&A, so please have some questions ready.

To kick things off, I would like each of you to give a brief introduction to your firm and strategy, and explain where AI fits in your process today. Divya, can I start with you?

Sure. It is wonderful to be here. Thanks for having us. It is great to see everyone.

We launched our firm, Avala Global, about [number unclear] years ago. We have about $2 billion in capital and about 25 people, roughly half of whom are on the investment team. It is exciting to start a firm in the midst of this explosion of technology and AI.

We are fundamental long/short equity investors. We always try to combine a top-down thematic lens on how the world is going to change in the coming decades with the bottom-up rigor of what is happening in our stocks here and now.

We think the opportunity is incredible in today’s market. The dispersion is very real, and, as I am sure we will discuss in a little bit, we are really excited about the internal use of AI too.

We have an incredible investor base, so thank you to everyone who is part of it: amazing endowments and foundations. It is amazing to keep growing with them. It is good to be here.

Thank you for having us, and thank you to GCM for being good partners. This is a pretty impressive conference you have put together. I have not seen anything like this from an individual allocator.

I am Soren Gandrud, CIO and Managing Partner of Jones Hill Capital. We are a good old-fashioned long/short healthcare equities firm, really focused on biotech.

We launched in January this year. So it is pretty timely [phrase unclear], because AI has been at the forefront of our thought process since we started putting the firm together in the middle of last year.

In terms of what we do, we really try to do three things. We focus on the parts of the market with the biggest dispersion, so we do a lot of biotech and we really like shorting stocks. We do deep, fundamental, bottom-up research and individual stock selection. And we put that together within a framework of strong, disciplined portfolio construction.

In terms of how we are using AI, it has been a bit of a journey. It started with: how do we get our data in place? It has slowly migrated to: what bespoke tools can we use to help expedite the investment process?

I have come from very large platforms that had amazing technology, but it has not been as usable from an individual investment-process standpoint. It has been much more helpful for understanding the risks of the portfolio.

A lot of the tools we are able to use now can be built exactly to our specifications. That is really exciting.

Soren, sticking with you: you launched in January and were building out in 2025, so you could integrate AI into your firm and process from day one. What were your priorities, and how did you approach that?

[Opening phrase unclear.] I knew this was something that was really important for us to put together. Literally from day one, from COO selection to every person we brought in, the business side of the firm has been aimed at: how do we do a better job, and how do we incorporate this new technology?

The learning process started with what drives it. To me, the most important thing is: what is our data, where does it come from, and how do we make sure it is really good? On top of high-quality data, there needs to be a judgment layer and frameworks that are bespoke to us.

It was really important to make sure that every piece of information we have in the firm was loaded into one place where AI could help us parse it effectively. That includes trading data, the big, high-level data sources, and the data we create on the team: how we value these companies, what our numbers are, and things like that.

You also partnered with an external vendor for some of that build-out. Could you tell us more about what mattered when you were looking at vendors, and what you saw in the vendor landscape?

We knew we could not screw up the data. I have to give Marty, our COO, a lot of credit for how we went out, sourced this, and interviewed a number of vendors. We found a really talented team called Hedgineer, who helped us early on with getting the data infrastructure rock solid and clean from the get-go.

We had the advantage of not having to rip out pipes and change the way data had been aggregated in the first place. They really helped make sure that foundation was rock solid.

After they had been there for a while, we realized technology was moving so fast that we needed our own internal resources. That started with hiring individual AI experts, but also transforming basically everybody on the business team into AI experts themselves. Everybody we hired, including our CFO and admins, has coding experience. They have gradually become part of the investment team and really helped us make these bespoke tools. The pace has been incredible.

Hedgineer, our original outsourced consultants, had been building tools guided by our investment process. Pretty quickly, we realized we could build much faster on our own. We went from a lead time of about a month last summer to having useful investment tools in our hands within days.

Divya, given your launch timing, you started at a different place, since you launched before AI became mainstream. How did you go about incorporating AI into an existing firm?

We are lucky that we launched in the past few years. We do not have very much legacy infrastructure, which really helped us move quickly and stay agile.

We launched in October 2022. Even before launch, which was before ChatGPT was announced, we could see from our work tracking these things that the end-market use cases for AI were picking up. We were able to put things in place very quickly that would let us use our data better in an AI world.

One of the key things has been the culture we tried to build. From day one, we tried to be very tech-centric. Part of that is that I grew up being very techy. I coded in a couple of languages in high school and was an engineering major. Seeing what OpenAI was doing back in early 2022 felt a little full circle for me, and I was excited about where the technology was going.

We invest in AI-related stocks. We were working in parallel on what AI might look like and the stocks involved in the AI build-out, while also trying to build it in-house. I feel lucky that we were able to build in this era before having a ton of legacy infrastructure.

What I mean by culture, and Soren talked about this, is that it is bottom-up. Every last person being bought into this vision is key. It is also top-down. I am passionate about it. I am using all these tools and building my own skills, and people see that.

We try to do things internally to enhance that culture. We do Friday hackathons [event nickname unclear], where we sit in a conference room for an hour or two and code or work in the terminal, creating skills. We share skills with each other and see what has the most uptake from analysts.

From early on, we tried to shift from just using chat windows to having everyone in the terminal. We hired external resources to help train everyone, from the CIO to our executive assistants, on how to use the technology, work in the terminal, and create a skill. That is where the creativity and innovation are.

About a year ago, we hired our lead engineer. We have hired our second, and I expect that team to double in the next year. I do not even know that you need a huge team, because the sheer productivity of these tools is incredible. If you empower your analysts and business leaders and have a couple of other resources, you can do a lot.

I have always felt that the cultural part, the top-down passion, and making sure the whole organization is very AI-forward in its thinking are the most important things.

For the lead AI engineer and the team you are building, how did you envision that role, and what kind of background were you looking for when hiring?

Our lead engineer had worked at other hedge funds and also has a classic engineering background. I liked that because product management is an important part of developing tools. Having someone who is familiar with the hedge fund process is helpful.

That said, the tools are so interesting at this point. You hear from the private labs themselves that the code is writing the code. When you are running a fundamental strategy, I do not think you need the most seasoned expert doing quantitative machine-learning research somewhere. You need people with high agency who have some basic skills and can run at the problems.

One thing that is nice about AI is that it lets you create a superpower for each individual. Everyone runs their process a little differently; every PM in the industry does it a little differently. Everyone has different superpowers, and AI lets you enhance your own, rather than there being one solution for everyone. Enabling the team you have is the most important thing.

Soren, similarly, when you were hiring at Jones Hill, what were you looking for in some of those roles?

First, we wanted someone who knew what they were doing and really understood the basic engineering perspective.

Similar to Divya, we hired a lead AI engineer who had both hedge fund and big-tech experience. She brought creativity that I think you do not necessarily get if you hire only from hedge funds. There are always different ways to approach things. She also had general familiarity with workflows, the parts of the investment process, and how they come together.

It was important to have someone who could coach the team on using these tools. One of the big things we struggled with culturally was that we had a way of doing things, and everybody was ingrained in that model. What we do should not change, but how we do it has to change dramatically.

There is a spectrum. Some people are very committed to the old way and do not trust AI, its judgments, or its outputs. If anything, I veer more toward that. Then there is the other side, which can be dangerous: someone who wants to use AI for everything, including judgment calls. You need to be somewhere in the middle.

For us, it is about using AI for the things we are already good at, where we can eyeball whether something is right or wrong, and saving our judgment for the other parts of the investment process.

You came up in a very different research environment from the one we live in today. Related to what you just said, what makes a good analyst now versus when you were coming up in the industry?

The bottleneck has changed. It used to be aggregating and analyzing data, finding all the pieces, building models, running the DCF, and analyzing data when it came out.

We always had very bottom-heavy teams. We hired bankers and had roles for them. We hired PhDs and had roles for them. Whether or not they had judgment, there was work to do. You needed judgment to get to the next level, but there was a role for people who did not have good judgment.

Unfortunately, that role is gone, because now the bottleneck is judgment and what you do with the data once you have it. That has changed how we evaluate analysts on the team, what we look for when bringing people in, and how we structure the business.

The most important parts of becoming an investor are even more important than before. As many things as AI can do, I have found it to be a remarkably terrible stock picker. Hopefully, we can now direct all the analysts’ mental energy toward that part of the process.

Divya, AI has changed a lot in its capabilities over the past year or two. What have you seen as you have incorporated it into your firm? As it helps you become more efficient, where do you think about spending your incremental time?

I completely agree with Soren. Judgment is what humans, and really strong portfolio managers and analysts, do best. AI lets you spend even more time on that because it reduces the time spent on things that do not need your time or attention.

I see it impacting every part of our investment process. In idea generation, it enables us to tackle a broader array of top-of-funnel ideas. We typically have a concentrated book, probably 20 longs and 40 shorts, but our universe is 2,000 to 3,000 names across most sectors globally, above a certain liquidity threshold.

Years ago, an analyst team might have tackled, just using round numbers, 100 names to dive into over the course of a year. With AI, we are finding that number is several times larger. You can very quickly assess the background of a company, the bull and bear cases, and the data behind them, using your pattern recognition and the way you like information laid out. The number of names we can get through and screen at the top of the funnel is several times what it used to be.

On the creative, deep-diligence side, it is also letting us go deeper. For example, there was an infrastructure software company we were interested in going long. Through a coding terminal [tool name unclear], we built a way to identify which AI-native tech companies were using it by tracking telemetry. We built a model of its customers and revenue base and forecast a huge acceleration, literally doubling from around 20 to 40 [units not specified].

That is the kind of thing we really could not do without going into the terminal and using it creatively.

On risk and portfolio management, it gives us stronger real-time insights that would have required much more manual work. At a high level, our vision of having agents covering stocks is almost like having many analysts covering these thousands of stocks, surfacing when something interesting happens. It focuses your attention on the most interesting parts of your universe, which is helpful when you have a broad universe.

A couple of years ago, a lot of people were just entering questions into a chatbot. We still do plenty of that, but now we are designing agent systems that run overnight and all the time, working on different things. Sometimes it is deep diligence. Sometimes it is tracking the background of a stock and telling you to pay attention when the right set of facts aligns. There are real-time insights on the risk side too.

With that broader funnel for idea generation, how do you prioritize now that so many more ideas are hitting your radar?

At the end of the day, I think it is resulting in two things. One is, hopefully, a more optimal portfolio. We are not suddenly becoming much more diversified, but because we can go through more top-of-funnel ideas, we have turned over more stones.

The other is that when there are market dislocations, we can do our work very quickly. At the end of March, for example, one of our big themes was AI, which we follow closely. We had seen a massive inflection around December 2025 and the first quarter of 2026 in coding tools and some of the private labs we invest in. Geopolitics gave the market a shock in March and created interesting entry valuations. We were able to run hard at many more names than we could have without AI. It lets us step in more during dislocations and capture them better.

You do have to design your systems correctly. You do not want a system that gives you a million flags or a bunch of noise. We are carefully calibrating everything as we go. You do not need 2,000 stock agents giving you [phrase unclear] on 2,000 stocks every day. That is not helpful.

As we design these tools and agents, we are iterating to fine-tune what is interesting enough to flag. It is a process, but at the end of it I think it will be really powerful.

Soren, what about Jones Hill? Have you used AI for idea generation as well, and how do you handle prioritization?

It is not that different. It really increases the number of names at the top of the funnel, and the hard part is deciding which ones to prioritize.

What helps us, even though it is biotech, is that we have a firm valuation construct for the different types of companies we invest in. When you have a big [market-move term unclear], which we just had in our sector, there is a historical tendency to put more capital into names where the fundamentals have not gone against you.

Now we have the data to understand whether the best opportunity is the thing we already own that has not worked, or whether it is another name in an adjacent area. How do we rank them? Who has the best catalyst pathway? Who has the best valuation? What are the highest-quality companies we can buy on sale?

That has been the most helpful part: organizing my thinking about what we know and helping the analysts organize theirs.

Given your focus on specialized healthcare sectors, particularly biotech, how has AI helped or not helped? Are there nuances to that research process, or places where you have found it particularly useful or not useful?

Every sector has nuances. [Brief comparison with Divya’s sector coverage unclear.] In biotech, there are two main verticals, with a lot of rabbit holes to run down.

One is valuation. AI has made it very easy to rank stocks and understand which are relatively expensive or cheap for what they are.

The second is clinical data, which usually requires the strongest judgment. Historically, our best analysts would take two weeks to work up clinical data. They would go scorched earth, downloading every trial the company had run on a particular drug, every other drug with a similar mechanism, and every other drug studied in the same indication, making sure they had found all the pieces. Then they would organize data that came in all these different forms.

Now AI can search the landscape and pull those pieces together. We can spend our time checking that the data and sources are right, making sure the information is complete for what we are looking for, and using AI to catch blind spots.

For the judgment call about whether a trial will work, we have a very specific framework. At the end of the day, though, it is a judgment call we have to make. AI is also not good at calling whether clinical trials will work. But it helps shorten that work from two weeks to more like two [unit missing in supplied transcript].

Are there other areas where human judgment remains essential? There is trial data, and I assume there are portfolio-construction elements.

All of it. Specifically on stock picking, as we discussed earlier, I joked with Divya that an old boss of mine thought most sectors were 70% fundamentals and 30% psychology, while biotech was the complete opposite.

Judgment still matters. It is the biggest thing we focus on. Taking the 80% of our time that used to go into gathering and parsing information and doing rote mechanical tasks, and reducing that to 10% or 15%, is really helpful.

It lets us focus our resources on judgment. It also lets us hire and evaluate people better, because everything else is standardized and their judgment becomes more of a differentiator.

Divya, you described a progression toward more agentic, multistep systems. Where are you today, and what hurdles have you faced?

We have had persistent agents running for a couple of months. The latest releases from the frontier labs have been amazing. They let you do so much more. Many of you play around with [OpenAI tool name unclear] or other things.

I walk in every morning, without even having explicitly programmed everything, and it tells me which models have been updated, whether there are any errors in the models, and what new evidence supports or does not support a thesis. It even helps keep you organized on non-stock-related things.

We are still in the early innings. The models are going to get much better at what they can intuit once they can see all of your company’s work. Going back to what we said earlier, we do not have legacy technology systems that have existed for 20 years. We have organized our data and built relatively AI-native systems. I think that will be a huge advantage as the models improve.

But you have to calibrate along the way. They do not do everything automatically. Investing has an art and a science. You can have the model do some of the science. Modeling is getting better and better, but there are many times when a really cool modeling [word unclear] will come from your creative thinking about how to analyze and model something better, rather than from a frontier-lab model.

They cannot go to a management meeting with a CFO you have known for 10 years and read her the way you can. They cannot go to a field conference in Kansas on a sector you care about and do that kind of work. You are constantly incorporating that real-world work into these agentic systems.

Judgment is important, and calibration is important. You build persistent agents, but you have to calibrate them daily. If I am building my Intel stock agent and go in and calibrate it every day, it will get pretty good pretty quickly. You cannot just put some things in and hope it becomes the perfect agent.

It takes time and focus. That is why the culture matters so much: every person in the firm needs to be bought in, because calibrating these agents takes time out of your day. If you do it well, it can be an incredible differentiator and let you do more with all the information out there.

For the agents you and the team have built, what does the verification and quality-control process look like?

At this point, there is nothing I simply trust as fully accurate because my agent gave it to me. A lot of it is agents flagging where you should spend time and making things happen faster.

We are nowhere near a place where they pick stocks. They are not good at that. You would never say, “It told me to buy this stock today; let me let that influence my decision-making.” That is not where they are at all. You have to verify.

We have an internal policy that if anyone uses AI for an analysis, they cite at the bottom of the work how they used it, so the PM and CIO are aware. There are many things you have to keep calibrating. We make it clear to the team that you cannot take something as fact just because it comes back to you. That is an important part of the culture of assessing and working with these agents.

Before we transition to Q&A, I have one or two final questions. Looking ahead, given how quickly this is moving, what capabilities or developments in your pipeline are you most excited about?

The big one for me is having something that can judge us and give us feedback on how I and the analysts pick stocks.

I have come from big places with amazing analytics. I would get binders about how I traded, but none of it was that useful because everything is different and context-dependent. It has much less to do with whether you buy or sell something more often when it is down, and much more to do with your rationale.

One of the things we have put in place, for me and all the analysts, is being specific about our theses, with clear ways we might be right or wrong and a list of what we are basing decisions on. We can collect that data and get real-time feedback on becoming better stock pickers and avoiding mistakes we might otherwise make over and over.

We try to mitigate blind spots and identify opportunities we should take advantage of, where there may be inertia. That is what I am most excited about. It will be a longer-term process, but we are already starting to get some of that data in. It is powerful, humbling in some ways, and really exciting.

I agree with that use case. To give an example, we recently did a month-long project on whether there are more optimal ways to manage risk, given that the market seems to be structurally getting more [word unclear] for a host of reasons.

Two or three years ago, our risk team would have done a lot of manual data crunching. It would have taken months to get through all our ideas. We had several hundred ideas or hypotheses to test, covering factors, positioning, and all sorts of portfolio-construction questions.

We were able to run several hundred backtests [tool name unclear]. We could iterate with the risk team almost every day or every other day, going through a few more backtests and analyzing them. We found five to seven insights that we are incorporating into our risk model. It took several hundred backtests to find those insights: things that would not have changed our historical returns, which we are happy with, but would have reduced volatility.

That is one example of using AI in the way Soren described. Overall, it can multiply the impact you have. If you look at a portfolio manager today, the percentage of the day spent truly judging a mosaic of information is probably a minority. There are many other things we do. I see a world where, pretty soon, judgment is the majority of our day.

For great PMs, that is their superpower: looking at a mosaic of information that they and the team have worked on and making the right judgment call. It is also what most portfolio managers and CIOs enjoy. I think the job becomes more fun.

It applies to every part of the process, from optimizing idea generation and doing more creative diligence to risk and portfolio analytics. You can almost scale your organization without necessarily adding people, because you have agents doing different things.

It takes time, dedication, and effort to build that. It tends to be a nights-and-weekends project because we all have our day jobs. But it is fun. One thing that has been cool to see is how fast the payback period is. Within months of hiring our lead AI engineer, we were seeing impacts on the book, whether in top-of-funnel idea generation or finding more ideas during dislocations.

The models are only getting better. Where could we be a year from now, in terms of people enjoying their jobs more, spending more time on their superpowers, and having an actual impact on the portfolio, returns, and our ability to monetize [final word unclear]?

Both of you have spent a lot of time and effort building this into your firms. What is your top piece of advice for managers who are earlier in the journey, whether on culture, hiring, selecting partners, or anything else in the build-out?

The quick answer is data infrastructure, because that drives everything. AI is only as good as the data going into it. Having something that is open-source and searching the internet can be very caustic to your investment process.

We spend the most time understanding our data sources, vetting them, and slowly rolling fully vetted new ones into our system, then continuing to assess and upgrade them where we can.

If you get the data right, many of the other pieces are easy to put together. Without that framework, it is garbage in, garbage out.

I completely agree on data. Organize your data first and get that right.

I would add the cultural element. AI is an incredible opportunity for everyone to be more creative. Keep encouraging that in every way you can.

When everyone is living and breathing it every day, the first thought becomes: “Is this the best way to do it, or is there a different way now with all these AI tools?” You suddenly create entirely new processes. I think that is very powerful.

Great. With that, we would love to open the floor for Q&A. Does anyone have any questions?

I am not entirely clear on this. You are saving a lot of time on research and information gathering, but are you training models, or using them just on the research side? If you are training, validating, and testing them, how are the test results comparing with the training results? [Question edited from a garbled passage; exact wording uncertain.]

We are not training any of the models. The labs do a better job than we do of building them. They turn out new ones all the time, and they are getting better to a startling degree.

Think of what we do as putting together bricks. We make sure the bricks of data we would normally look at as a firm to analyze a stock are assembled and validated. Those bricks then build into larger components: skills for how you normally do something.

Take something mundane like a DCF analysis. What sources are we using? Where are we pulling from our model or consensus? What assumptions do we make for a DCF analysis on a biotech stock? That becomes a skill. It used to take time to run. Now, instead of an analyst building it, it can be done quickly to our own specifications.

More of those pieces build over time into higher levels of complexity. Much of it is getting the data right, giving the model context, and letting it do its job within that.

I agree. We are not training models; we do not think that makes sense for our strategy. The frontier-lab models are incredible, and building systems that can easily work with the next model releases is important.

In the first half of the year, we did an analysis to start quantifying the impact, although I think it is early. We were able to tackle about four times as many top-of-funnel ideas as in the prior year. During dislocations, we were able to tackle even more, both in the number of ideas and the depth of work.

The data will become clearer over a longer period. The ability to create persistent agents is very recent, and we are all building right now. I have not built my 2,000th stock agent yet.

We are in an interesting moment. Between the strength of the models, including those released recently in the fall, and what we are building, we will be able to do a lot more in the coming year. I suspect we will say that every year. Running as hard as you can at this will be an incredible advantage, and I suspect we will be able to measure much more in the coming years.

Are you taking steps to introduce flexibility and reduce dependence on the frontier labs, perhaps through local inference?

For example, in biotech, over the summer, if you asked one of the Opus models for a bill-of-materials audit on a bioreactor, you got [response unclear] from Claude. Have you experimented with running inference locally or in the cloud so you are not so beholden to the frontier labs?

Fortunately, we are not yet trying to build our own agents of mass destruction.

What we try to do is trust that Claude or OpenAI will do a good job. Many of our tasks are, I will say, lower-level thinking: rote exercises. We have not had the need to get too creative on that front. [Brief aside unclear.]

We are letting them do what they do best. Much of that is organizing the data we put in, which lets us analyze it faster. As these things accelerate, it may be something we have to consider.

We are the same. We are thoughtful about which model we use for each task. We have checks and balances so someone does not accidentally run a crazy request on the highest-end model and leave us waking up to something that does not make sense.

It is clear which labs are at the forefront, and I want access to the best models. Longer term, a model one or two generations behind the state of the art will be very cost-effective. It already is, depending on the task, and we use those models for many things.

When I think about where we get the greatest return on our building time, it is less around that dynamic and more around the things we have discussed here. I do not worry too much about dependence. Given the pricing and the nature of the business model, you are not locked in. You can switch models very easily.

Thanks for a very interesting session. Could you bring to life how you can increase what goes into the funnel? You mentioned four times; perhaps you have a different number. How does that work without compromising quality? What do you trust the model to do, and what do you need to go deeper on yourselves?

I do not think we spend more time in aggregate at the top of the funnel. AI has shortened the process per stock so much that we can get through far more in the same amount of time.

The next step is filtering. We do what we call a two-day quick take. If it is a company we already know well, we may not need that; we are looking at the new facts. For a brand-new company, we do the quick take. This is probably a little different for us because we have a broader universe, whereas you probably know all your companies cold.

Then judgment comes in. You use the pattern recognition and judgment you have developed over decades as a PM or CIO to decide whether it should go to the next step and someone should do a deep dive.

We have not seen a greater time sink at the top of the funnel as it has expanded. On a per-stock basis, we can get through and gather so much more data and information to begin with. I do not see us sacrificing deep work. If anything, AI has let us go deeper and ask questions we never could before.

I mentioned the infrastructure software company, where we could have a differentiated view on a revenue-growth inflection. With data-center companies, we can scrape town-hall meeting notes and figure out where construction is ramping and where it is not. For our Asian stocks, we can look at Taiwanese stock [source type unclear] in a way we never could before.

I only see our work deepening, which is exciting. But prioritization and judgment become very important. In the coming decade, I think the CIOs, PMs, and teams with the best judgment will start to outshine those who may simply have had an advantage from a good system for going through alternative data. As AI can do more for you, judgment becomes more important at every step.

I will echo that. We spend much less time on lower-level tasks and much more on judgment. We are very micro-oriented and know the names in our universe.

One way we bring ideas to the top is to plan a lot in advance. We have something called a “strike zone” for every name in our coverage universe: an overlap of catalyst flow and valuation where, theoretically, we want to get involved. If a stock enters that strike zone in time or price, what are the few things we need to do in deep-dive research to get there?

In the past, it relied on Soren looking at the screen and something in one of my Launchpads jumping out at me. I spent a lot of time watching and remembering these mental constructs. We had them all planned out.

Now I come in every day and get 10 to 15 stocks that are within our strike zones. That is where judgment comes in: what are the one or two questions I need answered to feel comfortable, and which is the best opportunity?

Sometimes the answer is that we need to do 20 doctor calls on one subject to understand how a drug launch is going or what the key pieces are. But we are missing far fewer names when they enter those strike zones. We have other screening mechanisms too. It lets us prioritize better and lets fewer things slip through the cracks.

I completely agree. One more example happened the other day. There is a stock that, nine times out of 10, is a short when it gets to a certain level, and I want to short it. But it is not automatic; you have to look at the situation and context.

In a universe of thousands of stocks, before AI I might not have noticed that it reached that level. You could set flags, but then you would be going through a thousand flags on a couple of thousand stocks every morning.

Now, with agents, that will be surfaced. We call it something different from a strike zone, but it is the equivalent. You can capture more of those opportunities instead of leaving them unnoticed.

Great. With that, we are at time. I really appreciate you both taking the time to share your insights. AI has been hugely important, but judgment is still critical. [Final closing words unclear.]