Data integration connects applications, databases, devices and partners so that data reaches every system that needs it, in the right format, at the right time and under clear governance. This page collects my work on data integration architecture: the independent landscape, the architecture guides on event streaming, APIs and batch, legacy and ERP integration, and how integration provides the data foundation for process intelligence, workflow orchestration and AI agents.
Every integration platform has a design point, and the design point follows one of three communication paradigms.
Most enterprises use all three at the same time. What decides the quality of the architecture is which paradigm sits at the center. With event streaming as the backbone, request-response and batch become consumer interfaces at the edges, and an API, an analytics pipeline and an AI agent can all use the same data without coupling to each other or to the source systems. Event streaming earns that central role when several consumers need the same data, when latency affects business outcomes, or when AI agents move from pilot to production. A mid-size company with a mostly SaaS landscape and daily analytics is often well served without it.
A typical Global 2000 enterprise runs an iPaaS for API-led connectivity, an event streaming platform, a data management suite for batch and data quality, a message queue for transaction-critical messaging, and at least one legacy ESB. That mix is the result of decades of decisions, and replacing all of it with one product is rarely the right goal. The better goal is an integration architecture that connects the existing platforms, with event-driven principles at the center and governance treated as a first-class concern from the start.
The market is consolidating around suites that cover all three paradigms. IBM acquired Confluent, Salesforce acquired Informatica, Qlik absorbed Talend, and Fivetran merged with dbt Labs. Choosing one of these suites increasingly means committing to its ecosystem across API management, data integration and event streaming at once. Legacy formats shape the architecture at least as much as the vendor choice. SAP iDoc and BAPI, mainframe CICS, IBM MQ, EDI, SWIFT and HL7 stay in production for years, and the modernization path runs through change data capture and adapters into the streaming layer while the source systems keep running.
An AI agent that queries stale data does not just return a slow answer, it acts on the wrong one, which is why current, governed and continuously flowing data is the precondition for agentic AI in production. In an event-driven architecture, one streaming backbone serves three kinds of consumers: operational systems through streams, analytical systems through open table formats such as Apache Iceberg, and AI systems through context exposed via the Model Context Protocol (MCP).
Data integration moves and reshapes data. Workflow orchestration sits above it and coordinates what runs next, across data pipelines, infrastructure, applications and business processes, including retries, approvals and recovery. Integration middleware such as Kafka, ESBs and iPaaS stays in place underneath. Together with process intelligence and trusted agentic AI, data integration forms what I call the Trinity of modern data architecture.
This is how the topics on this site fit together. The Data Integration Landscape covers the three paradigms and their vendors. The Data Streaming Landscape zooms into the event streaming backbone. The Process Intelligence Landscape covers process mining and workflow orchestration on top of the data, and the trusted agentic AI and industrial IoT pages cover the AI agents and the plant floor that depend on it.
The map below places integration vendors in the three paradigms, request-response, event-driven and batch, ordered by integration scope.
The Data Integration Landscape 2026 maps API management, iPaaS, event streaming, change data capture, ETL and lakehouse vendors from the hyperscalers and enterprise suites to specialists and emerging players. It covers three generations of integration platforms, deployment models from SaaS to BYOC and edge, the legacy integration challenge, industrial IoT as a complementary world, and three architecture decisions that matter more than vendor selection. It is an independent practitioner perspective. No vendor pays to appear. Download the PDF or read the announcement post.
The event streaming layer is where most of the innovation in data movement happens, from diskless Kafka and Apache Iceberg to stream processing close to the source.
Most integration programs start with existing middleware. These posts compare event streaming with ESB, ETL, message queues, iPaaS and integration frameworks, and show how to combine and migrate them step by step.
Mainframes and ERP systems hold the most critical business data and stay in production for decades. These posts cover offloading, integration and gradual replacement without disrupting the core.
Looking for more? All posts in the Data Integration category list every article on the topic.
I advise enterprises, system integrators and investors on data integration architecture, including integration platform strategy across APIs, event streaming and batch, middleware modernization from legacy ESBs and iPaaS, the move to an event-driven backbone, mainframe and ERP integration, and the data foundation for AI agents. My career covered all three paradigms, with roles at Talend, TIBCO and Confluent. The advisory work is vendor-neutral and grounded in two decades of enterprise architecture across more than a dozen industries. Work with me describes the formats, from architecture reviews to workshops and keynotes.