Skip to main content

2026 & Beyond: Agentic Future in Finance

Summary

  • Three critical questions: This video frames the agentic future around three essential questions: who will win the AI model race, what risks must be managed, and what defines competitive edge in AI-powered financial services.
  • Contextual intelligence advantage: The key insight is that organizations building AI grounded in secure, governed, and contextual enterprise data will lead. Context matters more than model selection on the Databricks Data and AI platform.
  • Financial services focus: Over 2,000 financial institutions rely on Databricks for AI agents and data. This video covers competitive advantage, risk management, and how five pillars enable delivering enterprise AI in finance.

2026 & Beyond: Agentic Future in Finance

Watch: 2026 & Beyond: Agentic Future in Finance
Junta Nakai explores the “agentic future” of financial services, showing why success with AI goes beyond choosing the best model. Instead, it hinges on Contextual intelligence—the ability to harness and activate proprietary data to create real advantage. He frames the future around three critical questions: who will win the AI race, what risks must be managed, and what truly defines competitive edge. The key insight is clear: organizations that build AI that understand your business—grounded in secure, governed, and contextual enterprise data—will lead. Databricks enables this by unifying data and AI, turning insights into meaningful, business-specific outcomes.
Chapters-

Chapters

FAQs

Why is contextual intelligence critical for financial AI?

Financial institutions have massive proprietary datasets that public AI models cannot access. By combining large language models with your secure enterprise data, you create AI systems that understand your business and make better decisions than generic models.

How does data infrastructure impact the AI model race?

While new models appear constantly, data infrastructure is equally or more important. Organizations with robust data governance on the Databricks Data and AI platform can quickly adapt to new models while maintaining business continuity.

What competitive advantages do financial institutions gain from agentic AI?

Agentic AI enables automation of complex workflows, fraud detection, risk assessment, and customer intelligence. Organizations that integrate these agents into business processes gain significant competitive advantage and improved profitability.

How do you manage risks when deploying agentic AI in finance?

This video emphasizes that success requires not just technology but also robust governance, compliance frameworks, and human oversight. The Databricks Data and AI platform supports these requirements through built-in governance and audit capabilities.

Full transcript

0:00 The Agentic Future in Financial Services

Hello. I head financial services at Databricks. Thank you so much for joining us today. Where we're going to talk about the agentic future in financial services. So, before we begin, I just wanted to give you a brief update on where we are as a business in financial services in Databricks today. And we're super happy to tell you that over 2,000 financial service institutions around the world currently rely on Databricks for their AI agents and data.
Nine of the top 10 G-SIBs in the world outside of China operate on our platform to today across a multitude of different use cases from risk to finance to fraud detection. We've become the central data and AI platform for many of the largest and most trusted organizations in the world. So, as we think about an agentic future, I wanted to spend the time talking about where I think we're going. And where I think we're going is nobody knows what the exact path is going to be, but I do think there are three things that all of us need to be asking, which is who's going to win the AI application or the AI model layer?
What are all the things that could go wrong as we prepare for a agentic bank, agentic insurance company, so on and so forth? And what is going to be my competitive advantage for my company in this world going forward? Right? So, who's going to win the model race? What are all the things that can go wrong? And what is my core competitive advantage? Those are three things I think all of us need to ask as we go forward.
So, who's going to win the AI model race? The quick answer is I don't know, okay? It seems to the best models in the world seem to change every month, every week. And there's incredible innovation that's happening there, but I do know three things, okay? One is that whoever wins the AI application later, you're going to need lots of GPUs, you're going to need lots of memory, okay? And most importantly, I think you're going to need a lot of data infrastructure.

1:43 Three Critical Questions for AI Strategy

So, in many ways, the data decision becomes a more consequential decision than the model decision in the future. And the reason I believe this is that the models today are very, very smart. They're smart enough. But the challenge they have is context, okay? You as an enterprise, as a bank or insurance company or asset manager, have all this proprietary data that the AI models today cannot access securely. And if you can bring those two things together, then I think you add tremendous value up and above and beyond what is possible or thought to be possible today.
I think the second thing that we need to talk about is what are all the things that can go wrong, okay? So, as coding becomes more prevalent across your organization, there's inevitably going to be agentic sprawl at your company, right? There's going to be if there's going to be 10 agents per person times the number of people that you've employed, it becomes extremely difficult to govern and figure out all the things that are happening across your organization and ensure that things are secure, compliant, and follow all the protocols that you've set up, okay?
So, in many cases, the democratization, I think, of coding and coding can lead to a sprawl and a data and AI mess unless we're careful about how we approach that, okay? The third thing all of us need to think about is what is my moat? >> other day. It's called, k a r l, where we took a uh open-source model, tuned it, and were able to achieve uh significantly better results at 47% less cost and 33% less latency. Right?
So, as you kind of think about the battle between proprietary frontier models, open source models, I don't know who's going to win um eventually, but bringing those models to your data is going to become a significant advantage going forward. And the 1% that you see on the right is simply saying less than 1% of a enterprise's data is publicly available on the internet today. Okay? So, meaning the models that are being trained on synthetic and internet data, by definition, do not have in knowledge or contextual information about the things that are proprietary to you.
And I think that's an incredible thing important thing for us to think about as we think about what is the competitive moat going forward. Okay? So, um again, uh what is my moat? Uh what is all the things that could possibly go wrong, and what's the innovation in AI models, and how is that going to impact my business? Okay? With those three questions in mind, I think there is a fundamental truth emerging, which is that increasingly, AI with general knowledge is simply a commodity.
Okay? So, if you have access to the latest frontier model, so does every single one of your competitors. So, how are you using the models, these powerful models, to create a unique advantage for yourself? That is the trillion-dollar question every financial service organization on Earth is asking themselves today. Okay, so I want to ask you a quiz. It says quiz two, but it's actually the first quiz, quiz one. Which is which one of these bankers are the agent?
Is it the banker on the left or the banker on the right? And this is a trick question in many ways, because the answer is they are both agents. Okay? One happens to be human, the other happens to be AI. Okay? So, as you think about like agents and what that means, I think the world is the paradigm is shifting. So, for the last 60 years, it's all it was all about FinTech eating finance. There's 30,000 new new financial technologies companies that are coming out there who are going after the immediate parts of an incumbent's business, and that was one of the defining trends of the last 6 years or 7 years as people worried about what's going to happen in the future.

5:31 Shifting Paradigms: Human and AI Agents

I think going forward, the opposite is true, which is that finance is eating finance itself. What do I mean by that? As I as I mentioned before, the vast majority of the data that's available for your company is not publicly available in the internet, okay? It is private to you. So, the companies that can leverage the data in the contextual information they have for themselves to the best of their abilities is going to be the ones that gain a disproportionate share of the value and the disproportionate share of the competitive environment that out there, right?
So, if you can take advantage of the contextual information that you have at your business, then that's how you start to build a moat going forward. So, the bank of the future or the insurance company of the future or the asset manager of the future are going to be those companies that can not only agentify many of the workflows that exist today, but also power those agentic workflows with the live and the proprietary data that is uniquely available to you so that you can use the value of the information that you have at the at the greatest level, okay?
So, at the core, Databricks is a system that collects enterprise data from all of your different systems, okay? It could be Salesforce, Workday, it could be any of these trading tools that you have, so on and so forth. Centralizes into a single place in the lakehouse and secures it. And from that point on, those are the data sets that feed these AI agents, these models, and these analytics, and that's what powers real differentiated business decisions for your business going forward.
Okay, that is the core of Databricks. And the main point there is that AI without governance is simply risk. Okay? As you think about what all of this means, getting all your data into a single place and providing with the tools is incredibly important, but without governing the data and ensuring that the people get right access to the right information becomes paramount. Okay? So, quiz number two. Okay? One thing I want you to all think about is if AI disappeared completely for the world tomorrow, what would the revenue impact on your business be?

8:12 Assessing the Real Impact of AI on Revenue

Okay, so if AI disappeared completely tomorrow, what is the revenue impact on your business? Okay? When I ask this question around the world, the vast majority of the answers I get is something close to zero. They will say, "Yes, AI's helping us with efficiency. I think it's helping us with productivity." But in 90% of the in the cases, is it really adding revenue to our business? And I think the answer largely today is no.
Okay? So, why is it no? As I mentioned before, you need AI with contextual information. In other words, AI with your context is your advantage. Okay? General models, frontier models, using all that is great. It might help you do something 20 or 30% faster, but is it really going to help you make critical proprietary live business decisions for you? Um that's a big question that all of us uh we need to ask. So, there's a lot of conversations today about the models, frontier models, these open-source models.

9:22 Building Competitive Advantage with Data Context

Um they are incredible models, right? They're getting close to um AGI in the sense they are incredibly smart today. But I think choosing the digital brain or choosing the model is actually the last decision that enterprises need to make, okay? The first thing is it's instead of going from right to left, I think the right way to think about it is perhaps left to right, okay? Which is what is the data that we need to feed these machines in order to make it smart and add value for us.
How do we govern it, right? What are all the things that we're going to do with this the data and technology? And then ultimately then you could add a reasoning engine that brings this all together, right? When you think about agentic systems, this is what how an agentic system is built. It is not necessarily just the model decision. Whereas I think many executives or more of the business people in the market tend to think of things right to left, but as technologists, as the future is being built, the fundamental winners of tomorrow's are going to be the ones who think more left to right than right to left.
So, what does competitive advantage look like? Right, again, it's the context that you have, the proprietary data you have, and your ability to govern data, and that's what powers agentic systems to be safe and proprietary and value-adding. And this is happening across the world as we see it today. RBC Capital Markets has built a equity research augmentation agent that really helps their research analysts write reports at a fraction of the time at 99% accuracy where they can go out and talk to businesses and talk to customers quicker than anyone else, right?
Which adds to their bottom line and their top line because then they now have a better differentiated products for their customers. There are people like MasterCard who are really imagining reimagining what it means to be a transaction business or a credit card business, right? So, instead of just focusing on transaction processing, the biggest growth for MasterCard today comes from intelligent services that they can add to their proprietary to their real-time data and sell that to their customers, right?
This is a business that's growing substantially faster than the rest of their businesses, right? So, this is being done. So, if you think about the differentiation of AI going forward, it's going to rely on three things, okay? Or it's kind of a maturity curve of three things. I think first most people start with operational productivity, right? What are the things that we could do to augment specific tasks and role tasks and roles.
And they move to workflows, right? How do we automate workflows? How do we do things differently, right? The RBC example is a good one that I just talked about before. Then ultimately the value truly comes from reimagining a business and what it is to use the data and AI for productive and revenue generating purposes, like I showed with MasterCard just now very briefly. All these are available on our website for customer stories if you want to do a deep dive.
So, let's go back to those three questions, right? Like what could go wrong? Who wins the AI applications and what is my competitive advantage going forward, okay? So, who wins in the AI app? I don't know, okay? But I do know that if you do if you have a single interface with a contextual information where you can use leverage any of the frontier models, the most advanced cloud models, open AI models, so on and so forth within the context of your own data, that's how you're going to win.
That's an example of the power of the workflows that Databricks provides is an ability to do all that in a single pane of glass, right? What can go wrong? Again, lots of things can go wrong, but you need to have a platform that can govern all your data, all your models, and all your agents as well, okay? This is the cornerstone of what Databricks provides to our customers through services and products like Unity Catalog.
And what is your advantage? Your advantage is really it's your AI's ability to know your business and use the proprietary information your special sauce and use it for productive purposes. That is your competitive advantage going forward. Again, this is what data bricks is at the highest level. Again, models are smart. What they need is more context. And if you bring all the data in contact to a single place, govern it, and then power your AI, that's how you start to build competitive differentiation and the modes around your business as the use of AI proliferates all around us.

13:54 Hypothetical Case Study: Detecting Fraud at Enron

So, uh as I continue I wanted to share share a little story. Okay? And this is kind of a theoretical made-up story, but I wanted to ask the question of okay, what does this look like in practice? Okay? What if smart AI had access to our data that only we had access to, right? And it could reason against that data. So, having a frontier model is great, but what if those models can come into your environment, understand your data, connect to your system all in a governed, real-time, and proprietary way?
What would happen? And you know, fraud detection is one of the biggest use cases that happen on Data Bricks today. And if you recall about two decades or two or three decades ago, um one of the largest fraud instances uh in the US economy happened, which was a company called Enron. Okay? Enron was a big energy trading uh company, and it turns out they were uh cooking their books. Okay? And as a result of that, in inside a fraud, their stock price plummeted and went to zero at cascading effects.
Uh this is what led to the Sarbanes-Oxley regulation in the US as an example, okay? So, what we wanted to ask ourselves is well, what if Enron had a platform like Data Bricks, right? What if they can reason on their own data as this or before this was happening? And this is incredibly important because actually there was a woman at Enron named, which is on the right, who was one of the original whistleblowers at Enron, right?
She was looking at the number numbers and she said, "Hey, something's not adding up here." But, she couldn't really prove it, okay? The information was siloed, you know, the people were were like, you know, guarding the information, so on and so forth. So, she couldn't really prove what was happening. But, now all this is publicly available in court documents and synthetic data and so on and so forth. So, we asked a hypothetical scenario of what if data and AI was democratized through Databricks at a company like Enron when people try to figure out what the fraudulent systems were, okay?
So, one of the incredible things about Databricks, as I mentioned, is the ability to take real-time data from all your different sources and put it into one place, right? So, to be able to see what's happening, and when you bring in all these different systems, in this case, the internal valuation uh system and the ledger, which is publicly uh reported outwards, you can see there was something weird happening, which is the value of Enron's portfolio was remaining flat, okay, in the blue line, yet their internal models were saying that the value of it was diminishing rapidly, right?
So, there was a big gap between what the ledger said something was worth and what their internal systems were saying some of these uh trading positions were worth, right? That's a big red flag. So, when you kind of think about this and start interrogating that data and reasoning on that proprietary data, one of the questions we asked is, "Well, looking at this portfolio, why is it that the portfolio manager is going down so much where uh the portfolio value that we we report to the street is staying the same, whereas that's happening.
Meaning, the market value is going down, yes, but it turns out we have a hedge that was one-to-one inversely correlated with that, okay? So, it didn't we do we didn't lose any money because we have this perfect hedge. Okay, that's kind of, you know, that's kind of cool. Um what was in that hedge? Well, it was kind of a total return swap and it was the exact amount of money uh that we were losing. Okay, well, that's, you know, a coincidence and uh you know, we're fortunate it is.
That's great that we had great hedgers. So, let's double click on this a little bit more. Okay? You know, now I'm looking at different portfolios within Enron and now that I see that their special purpose vehicles, right? The equity in the special purpose vehicles is plummeting, right? And there's this rule accounting rule in the US that once you did be below a certain percent, okay? Uh you have to bring these offshore vehicles onto your balance sheet, okay?
And what's happening here is that the equity value of these special purpose vehicles were declining significantly. So, there was a risk that Enron had to bring all these off balance sheet items onto their balance sheet, right? Which would be terrible for the company. Um but from February to March, all of a sudden, the equity value jumps. Okay? So, now for some reason, all the special purpose vehicles are now really well funded.
Okay, that's kind of, you know, suspicious. And why is that happening? Well, when you kind of look at what who changed the models, it turns out there was a change in NAV calculation methodology between February and April, right? Someone came in and changed the model and how these things are calculated from something called mark-to-market to a model called hardcoded assumptions. Well, that's very suspicious, right? Um and when you kind of look through Uni catalog and because it's all government, you could ask the question, well, who changed the calculation methodology?
And it says the office of the CFO changed the calculation methodology. That is a big, big red flag, right? And, you know, because it's access through our proprietary systems and this is a true a real email uh from Enron, you could go through and say, "Hey, you know, theoretically, what were people saying about this?" Right? And it turns out that one of the most important people in the valuation team was saying that my group cannot support the valuation these things are being spitting out.
Right? So, even the people who are running the valuation groups were saying, "I feel strongly that I cannot support the valuation my group is producing." Right? So, again, if you have an incredibly uh a powerful model, which is great. But, when you can connect that to the contextual private information that is only available to you, the value of the agentic system, so the value of the AI come becomes exponentially higher, okay?
And here is an example of when you bring the context to the data and the AI to the data, the value, okay, of how these AI systems can work, okay? So, wrapping up, how do I win? Okay? We think there's five pillars of delivering true enterprise AI. I'm not going to go through each one of them, but I think the most important one for all of us to focus on is that we must start with the business challenge and not the technology.

20:07 Five Pillars for Delivering Enterprise

Okay? Like, too often I see people choosing a model and then figuring out out, you know, afterwards. It should be the other way around, okay? And for that, our teams here at Databricks has looked at the 2,000-plus customers that we have in financial service around the world. What are the specific use cases each and one of every one of them are doing so that we can help you build these use cases as quickly as possible, okay?
So, from the buy-side to insurance use cases, PNC, from banks, we've learned the patterns from you know, the thousands of customers that we have around the world to help you do these use cases as quickly as possible, and we have summarized many of these insights into something called the financial service outlook for 2026. You can scan the QR code here. And you start reading about all the insights, all the things that we've learned from our 2000 plus companies around the world in financial services about what your peers are doing.
Okay. So, finally, again, three questions all of us need to ask as we prepare for this agentic future. Who's going to win the AI application? What's going to go wrong? And what is my competitive advantage? Okay. And I said before, you know, there are many and it's almost impossible to figure out who's going to win the model race. Okay. There's many things can go wrong as many people start to build agents and you know, you have agentic sprawl.
And then the what is my advantage if everybody has access to these powerful models? Well, I think that the answer to all of these is that when you bring context to your AI, that is your competitive advantage, right? AI with your context is your advantage. Okay. And this is precisely why Databricks exists. Okay. The context is locked inside of your organization and Databricks helps to unlock it so you could help build that agentic future.
Thank you so much for joining me.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.