From insight to action: how LTAP could reshape the data foundation for the agentic era

For years, organisations have pursued a deceptively simple ambition: bring their data together to create more value. 

The data lakehouse marked a major step forward. It combined the flexibility and scale of data lakes with warehouse-style management and performance. Structured enterprise data could sit alongside harder-to-manage sources, including documents, images, video and other unstructured information. 

Yet one important part of the enterprise data landscape remains harder to bring into the same picture: transactional data. 

This data flows constantly from the applications and systems that run the business. Every card payment, retail transaction, inventory update or operational interaction creates new data. 

This applies far beyond retail. Industrial equipment generates telemetry, while aircraft and energy infrastructure produce constantly changing operational data. 

The same principle applies to operational information within a health care environment. This is data that tells us what is happening now, and in the age of agentic AI, that distinction is becoming increasingly important. 

Closing the gap between operational data and analytics 

Organisations have traditionally treated transactional and analytical workloads differently – and for good reason. 

Operational databases process large numbers of individual transactions quickly and reliably. Analytical platforms, meanwhile, query and analyse large volumes of information. 

Connecting them often requires change data capture, replication and transformation pipelines. These pipelines move operational data before it becomes available for analytics. 

Databricks’ Lake Transactional/Analytical Processing (LTAP) architecture proposes a different model. LTAP seeks to unify transactional and analytical data at the storage layer. Lakebase, Databricks’ managed PostgreSQL database, supports transactional workloads. The Lakehouse can then perform analytical processing against the same underlying data foundation. 

The aim is to reduce the duplication and pipeline infrastructure needed to keep environments synchronised. That matters because pipelines create more than technology complexity – they introduce time between an event taking place and that event becoming available for analysis. They also add components that organisations must operate, govern and reconcile. 

Reducing that separation could expand what organisations can do with their data. 

Why this matters for AI agents 

This is where the LTAP proposition becomes particularly interesting.  

Business intelligence has traditionally been relatively tolerant of latency. A human analysing a report might be perfectly comfortable working with data that is minutes, hours or sometimes a day old. 

An AI agent operating within a business process is different. 

If we expect an agent not simply to analyse information, but to reason and act, it increasingly needs access to the current state of the business as well as its historical context. 

Consider a retail agent responding to an unexpected change in demand. To recommend or initiate the right action, it could need historical sales patterns and customer information, but also the latest orders, stock positions and transactions. 

Or consider industrial operations. An intelligent system might combine historical maintenance information with the latest operational data from equipment to identify an emerging problem and determine the appropriate next action. 

The opportunity is therefore bigger than faster reporting. It is about reducing the distance between data, intelligence and action. 

Independent analysis of the LTAP announcement has highlighted the same challenge. The conventional separation between operational and analytical systems becomes increasingly problematic when AI agents need to reason continuously using both live operational information and historical context. 

That makes LTAP particularly relevant to what could become a defining infrastructure question for agentic AI: what data foundation do we need when machines can analyse and act at digital speed? 

Completing the enterprise data picture 

This represents the next stage in the evolution of the lakehouse. 

Initially, the lakehouse presented the ability to bring structured and unstructured information together on a common data platform. LTAP extends that thinking towards operational and transactional data: the information continuously being created by the applications and processes running the enterprise. 

That creates the potential for a richer data foundation for AI. 

For retailers, that might mean bringing customer, product and historical data closer to live orders and transactions. In healthcare, the opportunity could involve combining longitudinal information with appropriate real-time operational data. Across manufacturing, transport and energy, historical data could be analysed alongside the changing state of assets and infrastructure. 

Not every use case requires immediate data, and not every existing transactional architecture needs replacing. But where the value of a decision falls rapidly as data becomes stale, bringing the operational and analytical worlds closer together becomes increasingly significant. 

From architectural promise to enterprise reality 

There is, however, an important distinction between an exciting technology architecture and a successful enterprise implementation. 

LTAP does not remove the need to think carefully about data architecture. In many ways, it makes those decisions more important. 

Organisations still need to understand where their data originates, which system should own a particular write, how information should be governed, what level of freshness a use case genuinely requires, and how an AI system should be allowed to interact with operational processes. 

Some elements of the emerging LTAP architecture are also at different stages of product maturity, so organsiations need to assess capabilities against their own cloud environment, performance requirements and risk profile. 

This is where experience in enterprise data transformation matters. 

At CGI, we have been helping clients design, integrate and operate complex data environments for many years. We understand the data pipelines, legacy systems and operational processes that sit behind today's enterprise architectures – and the governance, security and organisational considerations involved in changing them. 

Our alliance with Databricks gives us an opportunity to combine that experience with an emerging generation of data and AI capabilities. 

The goal should not be to adopt LTAP because it is new. Instead, organisations should identify where closer alignment can remove genuine complexity or unlock new use cases. 

A new foundation for the agentic era 

We are still relatively early in the shift towards agentic AI. But it is already forcing organisations to reconsider some long-held assumptions about enterprise data. 

When AI moves from answering questions to participating in processes, access to trusted, governed and sufficiently fresh data becomes fundamental. 

LTAP offers an intriguing answer to that challenge by bringing another major category of enterprise data closer to the lakehouse – and, ultimately, closer to AI. The question now is where that convergence could create the greatest value. 

As enterprise AI moves into its next phase, its success will increasingly depend on giving intelligent models the right data, at the right time, within an architecture organisations can trust. 

Find out more about our work with Databricks