Why Management in the Financial Sector Is Rejecting Off-the-Shelf Software and Creating Their Own Solutions — and What AI Has to Do with It

I would like to share some impressions from the World Economic Forum 2026 in Davos. I think they may be useful for IT professionals. As is well known, the WEF is an annual meeting of leaders from the world's largest financial companies. Larry Fink, the head of the world's largest investment fund, BlackRock, serves as a co-chair of the forum. But what is noteworthy this time is that many discussions — and the most interesting ones often took place in the corridors between official presentations — were built not so much around artificial intelligence (AI) itself, but around the prospects for the classical IT sector in connection with it.

Vibe Coding: Every Employee Is a Developer

The word "vibe coding," which just six months ago was perceived as a joking term for non-specialists writing code, sounded entirely different in Davos.

Let me say this right away: the financial world is a fairly closed system. Everyone fears that competitors will get ahead. The coolest "features" are shared publicly only when something even cooler has already emerged. Still, without delving into details, the common denominator of the conversations was the gradual abandonment of third-party software and a complete transition to in-house corporate development.

This thesis itself does not surprise me. At Otkritie Broker, at the height of the COVID-19 pandemic, we launched a remote Innovation Factory with an emphasis on maximizing the potential of corporate IT development. At the time, there was a lot of talk about blockchain, and we were actively implementing distributed ledger technology in the organization — I'll return to this a bit later. There were also early elements of AI — in short, it was clear where things were heading.

What do we have now? As far as I know, all leading banks, investment funds, and brokerage firms worldwide have already established AI factories. Within these structures, software is being created, and the providers of well-known neural networks are not standing still either. Not only are they increasingly transitioning to AI-assisted code creation, but the emphasis in their business model is shifting from earning a few dozen or hundred dollars per month from each individual for paid subscriptions to their services, toward focusing on the corporate sector. At Anthropic (the creator of Claude), for example, about 80% of the business is now B2B.

That is, such companies — providers of neural network solutions — are now capturing the market from the classical IT sector by offering their solutions. But this works for non-financial legal entities. For banks and brokers, the situation is different: they are now actively replacing virtually any third-party software through vibe coding.

I sometimes see skepticism toward vibe coding — that it supposedly doesn't work. I can confirm that it does work, but of course, the details matter.

The Crisis of Trust in Third-Party Software in Finance

Growing distrust of third-party software in the global financial industry is not a recent phenomenon. If I try to explain why, it is because off-the-shelf solutions do not fully cover the specifics — for example, premium brokerage, private equity (investments in shares of non-publicly traded companies), and I'm not even talking about private credit (the provision of loans by non-bank structures on individual terms, without listing debt instruments on the exchange), which, by the way, currently underpins all the key financing for all well-known neural networks worldwide.

After the US and EU tightened the screws on bank lending following the 2008-2009 crisis, private credit received a boost. It is clear that even without modern AI development, third-party software would still have been in use, but often it was still necessary to create in-house alternatives anyway, including because of the need to ensure 100% protection against information leakage. Who and where can provide such a guarantee for third-party software? But the question arises — is there such a guarantee for software written inside the company? Opinions on this varied among those discussing the issue in Davos.

The Financial Sector Was the First to Feel the Changes

Well, we in the financial sector have become pioneers in the large-scale application of vibe coding, although this is not widely discussed publicly. But a fact is a fact — and that means we are the ones who must answer for how to guarantee that there are no vulnerabilities in the software created this way.

Here's the situation. Vibe coding allows for huge savings on programming costs. Management at some banks and brokers is already rushing to increase the ROE (return on equity) metric, trying to look good in the eyes of business owners. But I would not rush. It is better to direct the freed-up resources toward implementing blockchain in all processes related to the creation of corporate software.

The Future: Decentralization and Security

Here's what we see. Code is created, then a human edits it. The output is a product that has something from AI and something from a human. And we must clearly understand that errors can occur in both. And we need to know where, at which stage they occurred, and be confident that no changes were later made to the recorded sequence of events. And this can be guaranteed by blockchain.

Furthermore, I am a proponent of creating corporate software not on the basis of public neural networks, but on the basis of a corporate large language model (LLM). Some will say — that's expensive! But we need a model that is trained on specific areas — for example, private credit or working with modern fintech innovations. And financial organizations are capable of that.

Why have banks, investment funds, and brokers worldwide begun abandoning public LLMs — something that was discussed extensively in the corridors of Davos? The fact is that there is no guarantee that information entered as a query will not leak sideways. And this is not just about indexing AI chats on the internet. It's clear that many neural networks offer an option to avoid this, but the information is still stored — and on whose servers and in which countries — that is not disclosed.

The same questions arise in connection with the popularization by several companies that manage well-known neural networks of AI agents that will take over operational processes in financial businesses, either independently or with minimal human oversight. But where is the control? Where is the guarantee that there will be no information leaks or AI hallucinations? And let's be honest: even the human overseeing AI agents can also make mistakes (accidentally or intentionally). So a system is needed where errors are caught before they can significantly impact the key parameter of any financial service — trust.

The main challenge is not writing code, but ensuring its stable operation, compliance with regulatory norms, and integration with other internal systems of the organization. This means that the entire process should be placed on blockchain, and code creation should take place within a corporate neural network.

The specifics of financial vibe coding also require increased attention to security. The fact is that one cannot create "clean" software and then feed data into it. The software itself is highly dependent on what data is entered into it — and finance is not just about numbers, but about a great deal of non-numerical information. The financial sector is critically dependent on sentiments, emotions, and even on what clients dream about. The success of a corporate strategy is the ability to anticipate dreams, to see what clients will want tomorrow in order to prepare today. The information used in developing and testing corporate neural networks and software has such high value that even the slightest probability of its leakage cannot be tolerated. This is why such emphasis is placed — and will continue to be placed — on creating proprietary solutions, especially since the technologies to make this happen now exist.