I was fortunate enough to be in New York last week for the latest in the series of MongoDB.local events. Alongside a welcome break from the unapologetically autumnal (i.e. cold and wet) weather in the UK, it was a great opportunity to hear the latest on the firm’s product strategy, latest engagements with financial institutions, and to catch up with friends.
The headline message from the event was Mongo’s growing ability to support enterprises in unlocking business value from their AI investments. It was announced that version 8.2 of MongoDB Atlas is now in general availability and will deliver improved query performance and scalability, as well as the encryption of data in use (in addition to at rest and in transit). In addition, the integration of the embedding and reranking models into Atlas from VoyageAI (which was acquired by Mongo in February) are expected to further enhance the ability of Atlas to support a range of GenAI and agentic use cases.
As you would expect, the underlying theme through the agenda was the importance of a modern data architecture to the success of an enterprise AI strategy. The principle of ‘garbage in, garbage out’ very much applies, and reflects the reality that the impact of any initiative built around GenAI or agentic will only be as good as the data the models and agents can access and be trained upon.
This is very much a lived experience for many in the industry of course and making that leap from lab to production continues to pose challenges. Indeed, 32% of banks surveyed in Celent’s 2025 Dimensions study reported that the inability of their current systems to support GenAI initiatives was a barrier to scaling their use of the technology. In other words, the ‘digital dirt’ accumulated over time through the combination of legacy applications (and their modification), introduction of new applications, and M&A remains a blocker.
This is not lost on MongoDB, and the biggest takeaway from the event was the formal launch of the Application Modernization Platform (AMP). The proposition is simple enough: AMP combines a series of AI-powered tools with small teams of consultants to migrate legacy applications into more modern scalable services built on top of MongoDB Atlas (and running in the cloud). The underlying concept being to help clients to unlock the value of the data held in these legacy applications and thereby support a range of AI-related use cases.
The approach has been steadily honed over several client engagements, including with Bendigo & Adelaide Bank and Lombard Odier. In both cases, the projects involved taking applications for which there was limited (or no) documentation or organisational understanding and then using the AMP tooling (a combination of LLMs and agents) to convert, test, and validate the original code in small sections. This approach was found to be the most effective to ensure that the functional output is equivalent to the original. Crucially, the use of LLMs and agents made the process several orders of magnitude quicker than through traditional methods.
Celent understands that MongoDB is in active discussions with several financial institutions about making use of AMP, highlighting the impact that is has already begun to have with this model.
Given the importance of addressing legacy applications to the future potential of LLMs and agents in the banking industry, this will be an area to watch.
