AI agents and enterprise applications are changing what organisations need from their data platforms. For CGI, Databricks is becoming a key foundation for this next-generation architecture: bringing trusted data, governance, business context, AI and applications closer together so clients can move from experimentation to production value.
AI agents are starting to change the role of the enterprise data platform.
For many years, these platforms were designed mainly to serve human users: analysts, engineers, BI teams, data scientists and decision-makers consuming dashboards, reports and models. Those users still matter, but agents and AI-enabled applications introduce a different set of requirements.
They do not just read data occasionally. They need to reason over business context, retrieve fresh information, respect permissions, manage cost, trigger workflows and operate safely inside business processes.
This is why Databricks should be central to the enterprise architecture conversation.
CGI's 2026 Voice of Our Clients shows that AI adoption is accelerating, but readiness is not keeping pace. Organisations are applying AI to core business and operational processes, yet many are still working through the data foundations, technology modernisation and workflow reengineering required to scale it safely.
That is the shift we are seeing in customer and peer conversations. Clients are moving from AI experimentation to enterprise agents and apps, and that puts new pressure on the data platform.
This is where Databricks becomes the key component. It cannot be treated only as the analytics platform behind dashboards. It needs to be modelled as the governed foundation that gives those agents and apps business context, fresh data, identity, cost control and accountability.
Why Databricks matters now
The first wave of generative AI often focused on experimentation: chat interfaces, productivity tools, proofs-of-concept and isolated assistants. Those experiments helped organisations understand what the technology can do.
Enterprise agents and AI applications raise a more demanding question. If an agent is helping a service team prioritise a case, supporting a finance team with forecasting, creating a customer briefing or triggering a workflow, it needs more than a prompt and a model.
It needs access to trusted business data. It needs to understand the definitions and context behind that data. It needs to know what it can use, what it can change and what should remain under human control. It needs monitoring, cost controls and an operating model around it.
This is where Databricks is moving beyond being seen only as the lakehouse behind analytics and reporting. Its direction is far more relevant to the next generation data platform: a governed data and AI foundation where analytics, AI, applications and agents can work from the same trusted architecture.
Without that kind of foundation, AI can remain trapped in demonstration mode. The experience may look impressive, but the organisation cannot confidently scale it into critical processes.
What an agent-ready Databricks architecture needs
A Databricks-led architecture for the agentic era needs several capabilities to come together.
It needs trusted business context, because enterprise advantage does not sit only in the data itself. It sits in the meaning around it: definitions, relationships, policies, ownership, quality expectations and domain knowledge. Agents need that context to produce useful outputs rather than generic responses.
It needs governed access and identity. Agents must work within the same trust boundaries as people and applications. They should only access the data, models and tools they are permitted to use, and those permissions should be understandable and auditable.
It also needs fresh and operational data where the use case requires it. Historical reporting data remains important, but many agentic applications need current state: recent transactions, customer interactions, operational events, case updates or application state. If the data is stale or hard to reach, the agent will either be limited or risky.
Finally, it needs cost visibility, monitoring and accountability. As agents become more involved in workflows, organisations need to understand what is being used, by whom, for what purpose and at what cost. They also need lineage, audit trails and timely intervention.
How Databricks enables the next generation platform
What makes Databricks significant is how these capabilities are coming together in one platform direction.
Unity Catalog provides the governance foundation across data and AI assets. Unity AI Gateway extends that control into model, tool and agent interactions, including runtime policies, observability and cost management. Genie and Genie Ontology point to the growing importance of trusted business context and natural-language access. Lakebase brings operational database capability into the Databricks architecture, supporting the application state and low-latency patterns that agents and AI applications often need. Databricks Apps provides a route to build and deploy data and AI applications directly on the platform. Agent Bricks and MLflow support the development, evaluation and monitoring discipline needed to move agents from demos into production.
The point is not that every client needs every capability on day one, rather that Databricks is aligning around a real enterprise problem: agents need trusted context, fresh data, governance and operational control to be useful in production.
That is why we see Databricks as more than a technology choice. Used well, it becomes a practical architecture for modernising data foundations and enabling the next generation of AI-enabled applications and agents.
What this means for clients
For clients, this changes the starting point.
It is tempting to begin with the most visible AI use case. That can create energy, but it can also hide the harder work. If data is fragmented, ownership is unclear, definitions differ, access controls are inconsistent, or the operating model is not ready, the use case will eventually expose those weaknesses.
That does not mean organisations should wait for a perfect data estate before starting. In most cases, that would only slow progress. The better approach is to connect practical AI use cases with targeted modernisation of the Databricks foundation underneath them.
In other words, start where business value is visible, but design the delivery so it improves the platform, governance and operating model at the same time.
This is particularly important for organisations with complex legacy estates. Valuable data may sit across on-premises platforms, cloud services, operational systems and sector-specific applications. Full modernisation may take time, but AI adoption is already creating pressure for faster access to trusted data.
How CGI is modelling this into delivery
At CGI, this is how we are shaping our Modern Data Pathway thinking and the accelerators that sit around it.
Our Databricks Brickbuilder Specializations in Public Sector and Generative AI reflect our experience helping organisations move beyond pilots to enterprise-scale deployment. The important point is not the recognition itself, but what it validates for clients: governed foundations, reusable accelerators, technical depth and delivery patterns that can turn Databricks capability into operational value.
The aim is to help clients move from AI ambition to Databricks-enabled delivery in a structured way. That starts with assessing data and AI readiness: the current estate, the quality of the data foundation, the maturity of governance and the use cases most likely to create value.
From there, the work is about modernising or migrating the priority foundations that matter most. That may mean moving selected workloads from legacy platforms, improving data engineering patterns, creating governed access to distributed data, or establishing the right Databricks architecture for analytics, AI and applications.
It also means designing governance and operating controls from the beginning. For agentic AI, governance cannot be treated as a document that sits outside delivery. It has to be part of how data, models, tools, applications and users interact.
Then comes the practical build work: selecting the first agent or AI application use cases, grounding them in trusted data, designing the workflow, testing outputs, measuring value and deciding what should scale.
Finally, clients need support beyond the first implementation. Adoption, optimisation, monitoring and managed operations become critical as AI moves into day-to-day work. A successful Databricks architecture is not only built; it is operated, improved and governed over time.
