AI & Agentic-ready Data Platforms: A Roadmap for 2026

Estimated reading time: 15 minutes

According to Gartner, by 2026, 80% of enterprises will have adopted a modern data platform architecture, primarily driven by the need to support advanced analytics and AI initiatives. The era of traditional Business Intelligence (BI) platforms as we know them is really coming to an end. As we move into 2025, more than supporting other use cases, organisations must reconceptualise their data platforms to support both AI and agentic systems. The days of BI-centric platforms are giving way to robust AI Data Platforms designed for real-time, intelligent operations.

This transformation is not just about technology – it represents a fundamental shift in how we think about, organise, and utilise data.

The End of Traditional BI-Centric Platforms

Traditional BI platforms, designed primarily for reporting and analytics, are no longer sufficient. In fact, BI is expected to represent less than 50% of data platform usage in the near future. The new principle requires platforms that can handle both structured and unstructured data in near real-time, with a strong emphasis on a centralised semantic layer and active data management and observability.

5 Pillars of Modern Data Platforms for AI & Agentic

The next chapters will dig into the five pillars of Modern Data Platforms for AI and agentic systems:

1. Centralised Semantic Layer: The Foundation for AI and Agents

The semantic layer has emerged as the foundation of Modern Data Platforms, particularly for AI agents and LLMs. The new semantic layer universally translates human intent and data structures, unlike traditional BI implementations that trapped semantic definitions within specific visualisation tools.

A robust AI data platform is the foundation for deploying sophisticated tools, such as the Enterprise AI Agents
detailed in our whitepaper.

Why Semantic Layer is Critical for AI Agents

AI agents, powered by LLMs, primarily interact through natural language. The semantic layer bridges the gap between these text-based interactions and the underlying data structures by:

• Providing context and relationships that make data meaningful to LLMs.
• Enabling natural language queries to be accurately translated into data operations.
• Maintaining consistent business definitions across all AI applications.
• Ensuring that agents understand and use data according to business rules.

Universal Data Access Through Semantic Understanding

The semantic layer exceeds its traditional role in BI to become a universal knowledge graph that connects data across domains, a central repository of business logic and metric definitions, an interpreter between natural language and technical implementations, and a guarantor of consistent data interpretation across all tools and platforms.

This universality consistently applies the same semantic understanding, business rules, and data relationships, whether a business user accesses data with natural language, an AI agent answers queries, an automated process makes decisions, or a data scientist builds models.

Breaking Free from Tool-Specific Semantics

Historically, embedding semantic definitions directly within specific tools like BI platforms created inconsistent interpretations across different systems. This approach led to duplicate definitions and business logic, hindered AI adoption and automation, and limited the ability to scale data understanding throughout the organisation.

The new semantic layer solves these challenges by centralising semantic definitions outside of any specific tool, making business context available as a service, enabling dynamic adaptation of data understanding, and supporting multi-modal data interpretation for AI systems.

2. Becoming Active: From Passive Storage to Operational Hub

The transformation from passive BI platforms to active operational Modern Data Platforms represents a fundamental shift in how organisations leverage their data assets. Traditional data lakes served primarily as static repositories for historical analysis, but modern platforms must become dynamic operational hubs that actively participate in business processes.

Beyond the Analytics-Only Paradigm

Traditional BI platforms were characterised by:

• One-way data flow (from operational systems to data warehouse)
• Batch-oriented processing
• Focus on historical analysis and reporting
• Limited interaction with operational systems

Modern active Data Platforms fundamentally transform data flow and processing by breaking the traditional paradigm. They enable bidirectional data flows, allowing for seamless interaction between different systems and applications. Additionally, they support real-time data processing, which facilitates immediate insights and decision-making.

Furthermore, modern active data lakes directly power operational decisions, providing the necessary data and analytics to inform business actions. Ultimately, they serve as a central nervous system for business operations, integrating and orchestrating various processes to drive efficiency and effectiveness.

Operational Capabilities and Requirements

The shift to operational AI Data Platforms demands:

Near Real-Time Processing

Streaming or near-real-time processing capabilities are essential for handling high-volume, high-velocity data streams in today’s operations. This enables the platform to process and analyse data as it arrives, allowing for timely insights and decision-making. Event-driven architectures can also help, as they enable the platform to react to specific events or changes in the data, triggering actions or notifications as needed. CDC or Data movement solutions (as Fivetran / HVR or Airbyte) can help bypass traditional ETL and batched operations.

Bi-directional Integration

Reverse ETL capabilities are essential for reinjecting data back into operational systems, ensuring that insights and analytics are generated and acted upon in real time. This capability enables the active data lake to consume data and influence operational processes directly.

An API-first architecture is critical because it directly facilitates interaction between your active data lake and various internal systems. This approach ensures you can easily access and integrate data and analytics into different applications and services. Ultimately, an API-first strategy fosters a culture of data-driven decision-making by standardising and securing data exchange.

Organisations need automated data-sharing mechanisms to streamline data exchange between different systems and applications. Automating this process allows organisations to reduce manual data transfer errors, increase efficiency, and ensure secure and controlled data sharing.

AI Data Platforms Operational Use Cases

Direct integration with customer-facing applications is a key use case for active data lakes. This integration allows for real-time data sharing and analysis, enabling businesses to provide personalised experiences to their customers.

Another important application is real-time personalisation engines. These engines use real-time data to tailor content and experiences to individual users, increasing engagement and satisfaction.

Dynamic pricing and inventory management are use cases that benefit from real-time data. By analysing market conditions and customer behaviour in real time, businesses can optimise pricing and inventory levels to maximise revenue and customer satisfaction.

Impact on Business & Technical Operations

This transformation enables immediate action on insights rather than delayed responses, automated decision-making based on real-time data, seamless integration between analytical and operational processes, dynamic business process optimisation, and real-time customer experience personalisation.

When making your Data Platforms more operational, you really must understand that your platform is becoming a real operational component of your Information Systems, requiring careful monitoring and operational surveillance, as any other active component in your IS. Traditional BI teams could not be used to this level of requirements :

• A lately delivered dashboard is harmful but not dangerous.
• An outdated customer scoring or segmentation could really harm your online sales.

3. Domain-Driven Architecture: Breaking Free from Traditional Constraints

The shift from traditional three-tier/medallion architectures (Bronze/Silver/Gold or Raw/Refined/Curated) to a domain-driven approach represents a fundamental rethinking of how we organise and manage enterprise Data Platforms. This transformation is essential for enabling AI and agentic systems at scale.

Limitations of Traditional Architectures

The traditional three-tier architecture creates several challenges:

Organisational Bottlenecks

Centralised teams are prone to becoming overwhelmed with requests, leading to a slow response to business needs. This is exacerbated by the limited domain expertise within these central teams, making it challenging to effectively address the unique needs of each business domain. Furthermore, the coordination between business and IT stakeholders becomes increasingly complex, hindering the ability to respond quickly to changing business requirements.

Modern Data Platforms Technical Constraints

The traditional architecture is characterised by rigid data models that are difficult to evolve in response to changing business needs. The complex dependencies between layers make it challenging to modify or update individual components without affecting the entire system. Additionally, the architecture’s inability to optimise for specific use cases results in a one-size-fits-all approach to data transformation, which can be inefficient and ineffective.

Scalability Issues

As the organisation grows, the central team resources scale linearly, leading to increased coordination overhead and difficulty in maintaining data quality at scale. Moreover, the traditional architecture’s limitations make it challenging to parallelise development, hindering the organisation’s ability to scale efficiently and effectively.

Domain-Driven Data Products Approach

The new approach organises data around business domains, where each domain:

Ownership and Autonomy

Clear domain ownership and accountability are essential, ensuring that each domain is responsible for its actions and outcomes. This autonomy allows for decision-making within domains to be self-sufficient, enabling local optimisation tailored to each domain’s specific needs. This approach directly aligns with business objectives, as each domain focuses on achieving its unique goals.

Data Product Thinking

Organisations treat data as a product, complete with clear Service Level Agreements (SLAs) that outline its quality, availability, and performance. Well-defined interfaces and contracts ensure seamless integration and consumption of data products. Built-in data quality and observability mechanisms guarantee the data product’s integrity and facilitate its monitoring. The primary focus is on understanding user needs and consumption patterns to ensure the data product meets its intended purpose.

AI Data Platforms Architecture Principles

This architecture empowers independent domain evolution, allowing each to progress at its own pace, free from hindering dependencies. We establish clear boundaries and responsibilities to prevent confusion or overlap. Standardised inter-domain communication ensures seamless interaction between domains, promoting a cohesive and integrated system. Implementing a federated governance model oversees the entire system, ensuring consistency and coordination across domains.

Enabling Scale Through Domain Architecture

This approach enables scaling by:

Parallel Development

Independent domain teams work autonomously, reducing coordination overhead and enabling faster iteration cycles. This autonomy enables domain-specific optimisation, tailoring each domain to its unique needs and objectives.

Clear Responsibilities

Domain teams manage their data products, ensuring alignment with specific domain needs. Central teams focus on developing platform capabilities, providing a foundation for the domains to build upon. Standardised interfaces between domains facilitate seamless integration, while a shared governance framework ensures consistency across the system.

Flexible Evolution

The domain-driven approach enables domains to evolve at different speeds, allowing for independent technology choices where appropriate. This flexibility enables experimentation within domains, fostering innovation and improvement. This also facilitates a gradual modernisation path, ensuring domains can adapt to changing requirements without disrupting the entire system.

Implementation Considerations

Success with domain-driven architecture requires:

Organisational Alignment

Establish clear domain boundaries and align them with your organisation’s business objectives to ensure each domain focuses on specific business needs and outcomes. Empowered domain teams are essential, as they are responsible for making decisions and taking actions within their respective domains. A balanced distribution of responsibilities between central and domain teams is crucial, ensuring that each team is accountable for its actions and outcomes. Establish strong communication channels to facilitate collaboration and coordination between teams, ensuring information flows seamlessly across domains.

Technical Standards and tooling

• To ensure seamless integration and data exchange between domains, shared data contracts must be defined and agreed upon.
• Standard integration patterns should be established to simplify the integration process and reduce complexity. Common quality metrics are necessary to ensure that data products meet the required standards across domains.
• A unified metadata management system is essential for maintaining a consistent understanding of data across the organisation.
• Modern Data Governance is no longer only about rules and PowerPoint but also robust tooling and platform involvement to help engineers and business users work seamlessly with Data.
• Data Platforms’ internal Marketplaces are becoming increasingly standard for packaging and delivering Data Products to internal stakeholders, as they could be for external users.

Governance Framework

A federated governance model is necessary to oversee the entire system, ensuring consistency and coordination across domains. Clear decision rights must be defined to avoid confusion or overlap between domains. Standardised quality measures are essential to ensure that data products meet the required standards across domains. Cross-domain coordination mechanisms must be established to facilitate collaboration and ensure that domains work together effectively.

This approach creates a more resilient, scalable, and agile data platform that can better support the diverse needs of modern enterprises, particularly in the context of AI and agentic systems deployment.

My point of view

I want to make it clear that the solution is not really in the adoption of any data modelling techniques (entity-relation model, star schema, snowflake schema, and other data vaults—which seems to bring today more problems than solving issues) but more in the data architecture reusing digital platform design, where pizza teams, APIs, and agility were foundational to bringing real time-to-value.

4. AI Data Platforms: From Data Quality to Data Observability

The increasing prevalence of near-real-time data and massive volumes has rendered traditional data quality approaches ineffective. To address these challenges, Modern Data Platforms must incorporate advanced capabilities that ensure data quality and integrity. Specifically, these platforms require end-to-end visibility across the entire data value chain, enabling the tracking of data from its source to its final destination. This visibility is crucial for identifying and addressing data quality issues promptly.

Automated quality testing and anomaly detection are also essential components of Modern Data Platforms. These features enable the identification of data quality issues in real-time, allowing for swift corrective action to be taken. Furthermore, advanced monitoring capabilities are necessary to proactively identify potential issues before they impact the data pipeline. This proactive approach ensures that data quality issues are addressed before they affect downstream applications or users.

Comprehensive lineage and impact analysis are also critical components of Modern Data Platforms. These capabilities provide a detailed understanding of how data flows through the system, enabling the identification of the root cause of data quality issues and their impact on downstream applications. This understanding is essential for making informed decisions about data quality and for optimising data processing workflows to ensure the highest levels of data integrity.

If you really want to know more about Data Observability, take a look at pure players like Sifflet Data (my favourite, yes, it’s French) or Monte Carlo Data.

5. Your unstructured Data is just … Data

In the era of AI and Agentic systems, the distinction between structured and unstructured data is becoming increasingly irrelevant. AI agents, with their advanced capabilities, can seamlessly process and analyse both types of data, extracting valuable insights and making informed decisions. Therefore, organisations should adopt a unified approach to data management, treating unstructured data with the same level of importance and rigour as structured data. The recent acquisition of Datavolo by Snowflake shows how platforms are preparing for this approach.

Why Unify Data Management?

AI agents are agnostic to data structures. They can leverage advanced techniques like Natural Language Processing (NLP) and Machine Learning (ML) to extract meaning and value from both structured and unstructured data. By unifying data management, organisations can unlock the full potential of their data assets, enabling AI agents to leverage all available information for decision-making.

How to Achieve Unified Data Management

• Unified Data Platforms: Organisations should leverage Modern Data Platforms that can seamlessly handle both structured and unstructured data. These platforms should provide a unified view of all data assets, regardless of their structure, enabling AI agents to access and analyse data from a single source.
• Unified Operating Models: The same security protocols, data governance policies, and operational processes should be applied to both structured and unstructured data. This ensures consistency and compliance across all data assets, regardless of their structure.
• Integrated Data Management: When using specialised components like vector databases for unstructured data, their management should be integrated into the overall data platform. This ensures that all data assets are managed in a coordinated and consistent manner.

By adopting a unified approach to data management, organisations can empower their AI agents to leverage the full spectrum of their data assets, driving innovation and unlocking new opportunities for growth.

Preparing for AI and Agentic Systems

Infrastructure Requirements

Universal Storage: The data platform of 2025 must efficiently handle both structured and unstructured data. This is essential to accommodate the diverse data types and sources that modern enterprises deal with.

For a structured format, 2025 will certainly see Iceberg becoming the new open de facto standard for multi/hybrid-cloud structured data storage.

Only advanced Data Platforms can guarantee a universal and always governed Data Storage for both structured and unstructured data.

Near Real-Time Processing: The ability to access and process data in near real-time is a critical requirement for the data platform of 2025. This capability enables organisations to make timely and informed decisions based on the most current data.

Semantic Understanding: The data platform of 2025 must possess advanced semantic understanding capabilities. This includes the ability to understand the context and relationships between different data elements. This understanding is crucial for enabling more intelligent and context-aware data processing.

Governance Evolution

The governance model is evolving from a centralised to a federated model, where:

The IT department is responsible for managing the platform infrastructure, ensuring its stability and scalability.

Individual domains are granted a high degree of autonomy, allowing them to make decisions that are best suited to their specific needs and objectives. Shared governance rules are established to ensure consistency and alignment across all domains, promoting a cohesive and integrated system. Mandatory data cataloguing and quality reporting are enforced, ensuring that all data products meet the required standards and are well-documented.

Clear data contracts are established between domains, ensuring that data is exchanged in a standardised and secure manner.

To ensure Modern Governance is not only rules and powerpoints, it must be enforced into the core of your data platform. This is called Federated Computational Governance, which includes :

• Standard as Code
• Policies as Code
• Automated tests
• Automated Monitoring

While a robust platform is essential, securing its inputs and outputs is paramount, especially with new technologies. Learn more about the specifics of Generative AI data governance.

AI Data Platforms: The Road Ahead

To sum up what we’ve just been through, organisations must focus on several key areas to prepare their Data Platforms for AI and agentic systems:

Industrialisation

• Complete CI/CD integration
• Automated infrastructure management
• Standardised deployment processes
• Federated Computational Governance

Data Synchronisation

• Move beyond traditional ETL
• Implement real-time data synchronisation
• Leverage cloud-native ingestion solutions

Accessibility

• Semantic layer within the Data Platform
• Implement “chat with your data” capabilities
• Enable natural language queries
• Support text-to-SQL functionality

A Transition To Modern Data Platforms

The transition to AI and agentic-ready Data Platforms represents a fundamental shift in enterprise data architecture. Success requires organisations to move beyond traditional BI-centric thinking and embrace a more dynamic, interconnected, and semantically rich approach to data management. The platforms of 2025 will not just store and analyse data – they will actively participate in the organisation’s AI and agentic ecosystem, enabling new levels of automation, insight, and innovation.