Just when you thought the noise around AI couldn’t get any louder, we got 2025…
Back in February I took a punt with five predictions for integration in 2025, fully expecting at least one of them to fail before the ink dried. Instead, we got a year that was busy, interesting, and in a few places surprisingly aligned with the forecast. This post is a look back at what actually happened in integration over 2025: where the industry moved, what quietly faded into the background, and how my predictions fared when they collided with reality.
The backdrop: integration stopped being optional
If there was a single constant theme this year, it was that integration was no longer treated as a side concern. Organisations added more applications, adopted more SaaS platforms and leaned more heavily on data services. Every AI initiative demanded clean and timely data flows. Every automation programme assumed systems could be stitched together on command. However, the skills required to do this safely and at scale remained scarce.
The result was a quiet shift in posture. Integration moved from background plumbing to something much closer to critical infrastructure. Leadership teams discussed API platforms and data integration in the same forums as ERP strategy. SaaS buyers assumed that products would ship with useful integrations already available. And integration teams found themselves balancing the day to day work of keeping the lights on with the strategic responsibility of shaping the organisation’s AI and automation roadmap.
It was against this backdrop that the defining trends of 2025 emerged.
From copilots to AI native and agentic integration
In my prediction piece I suggested that most new iPaaS features in 2025 would have AI written somewhere in the announcement. That proved true, but the real story was how the role of AI changed across the year.
Early in 2025 many platforms still focused on narrow copilots that generated snippets of logic or suggested connectors. By the end of the year, the leading vendors were talking about AI native integration and using the language of agents rather than assistants, while staying true to their individual platform strengths. Boomi pushed this shift with its low-code Agentstudio platform and a family of Boomi AI agents. Microsoft presented its Azure AI Foundry as an agent platform and extended Logic Apps to interact with it. MuleSoft continued to integrate more closely to Salesforce with its use of Agentforce as its agentic platform, leveraging the investments of its parent company.
The difference was (mostly) not just surface branding. AI was treated less as a helper and more as a new participant in the integration landscape that needed tools, permissions, context and observability. Integration teams were no longer only building flows for humans and applications; they were increasingly building the guardrails within which AI could operate safely. Understanding which problems can be solved probabilistically and which are deterministic is now a key ability for integration practitioners – especially as the AI hype volume is only increasing.
Verdict: Would you like AI with that?
The AI crash that never came
My second prediction was more cautious. With the amount of hype and investment swirling around AI, it seemed likely that 2025 would bring some kind of reality check. As it turned out, we got the opposite.
That did not happen. Investment in AI and GenAI continued to set records (worldwide spending on AI is forecast to total nearly $1.5 trillion in 2025), the major cloud and model providers increased their spend, and many enterprises treated 2025 as the year to double down rather than back away (91 per cent plan to increase [AI investment] this year). While there was much grumbling from executives about ungoverned copilots and proof of concept work that never made it into production, it didn’t seem to have any impact on AI spending either from platforms or consumers.
The integration market served as a perfect microcosm of this trend. In many releases, AI features effectively crowded out everything else, with improvements in areas like monitoring, transformation and connectors either reframed as AI enablers or quietly pushed back. Rather than a trough and a pullback, 2025 delivered record investment and an integration market that talked about AI more loudly than ever.
Verdict: Maybe next year?
Hyperautomation coming of age
Hyperautomation is not the kind of trend that generates dramatic headlines, yet it was a quiet driver of integration in 2025. Many organisations that had experimented with isolated RPA bots began stepping back and designing broader automation strategies. Rather than automating a screen or a task in isolation, they examined whole processes across finance, supply chain, customer service and case management.
As soon as they did that, integration surfaced as a bottleneck. Every system involved in those processes required a clean, reliable interface. In some cases existing APIs were insufficient or missing entirely, so integration platforms stepped in to fill the gap. Integration debt also became harder to ignore. The point to point scripts and temporary jobs that had worked for small teams began to break down under the weight of coordinated automation efforts.
At the same time, it became clear that RPA alone could not carry the load. Organisations that made the most progress used a combination of APIs, integration flows, event based triggers, desktop automation where necessary and increasingly, AI agents to handle unstructured input or decision steps. Integration sat at the centre of this mix, providing the fabric that connected these components into something coherent.
Verdict: RPA is dead, long live RPA Hyperautomation
Embedded iPaaS hit its stride
Somewhat of a stealth trend in 2025 was the rise of embedded integration within SaaS applications. The building blocks had been developing for years in the background. SaaS buyers wanted their products to integrate cleanly with core systems. SaaS vendors wanted to avoid the cost of building dozens of bespoke connectors. And embedded iPaaS providers had invested in platforms that could be woven directly into the product experience.
In 2025 this idea crossed a threshold. We saw more SaaS applications with integration marketplaces built directly into their user interfaces, with configuration screens that felt native even though they were powered by an embedded integration engine. Sales teams began positioning these integrations as central features rather than side benefits. The result was an ecosystem where embedded iPaaS became a standard part of the commercial and technical offering for many SaaS vendors.
Embedded iPaaS has moved out of the big iron ERP and into the modern SaaS platform. Enterprise integration tools are no longer required to deliver intra-application functionality, but are still the best solution for stitching together a SaaS-driven organisation.
Verdict: SAP PI redux
When integration, APIs and data finally moved in together
The most structural change in 2025 involved the consolidation of application integration, API management and data integration. In the prediction article I suggested that cloud data platforms were extending into integration territory and that integration vendors would respond by strengthening their data capabilities. That is precisely what we saw.
Perhaps the largest story in integration this year was when Salesforce moved to acquire Informatica to provide coverage of the data space where MuleSoft was traditionally weaker. Boomi continued to integrate its Rivery acquisition to create a combined story for data pipelines and traditional iPaaS. IBM pushed a unified hybrid integration narrative, combining the assets of webMethods and StreamSets.
The common thread was customer demand for a single place to define, manage and observe the connections between systems. Organisations wanted one catalogue for APIs, data objects, events and pipelines. They wanted one policy model and one set of governance controls. They wanted consistent lineage and monitoring across tools that historically lived in separate silos.
Unless you’ve been doing this for a while, in which case you’d say that the common thread was vendors either defending their patch or trying to muscle in on someone else’s. Take your pick.
Verdict: Salesforce Infor-mule-ica
Where this leaves integration heading into 2026
If 2024 was the year AI arrived everywhere and 2025 was the year integration absorbed the impact, then 2026 looks like the year where control becomes the priority. AI agents will need guardrails that are as robust as those applied to human users. Integration platforms will increasingly act as the control plane through which AI interacts with enterprise systems. Consolidation will likely continue as the boundaries between API, data and integration platforms blur even further. Event driven and streaming based designs will become more common as real time data becomes the default expectation for AI and analytics workloads. And integration professionals will continue to expand their skillsets as the discipline evolves.
What has not changed is the core value of good integration. Connecting systems, data and processes in a way that is reliable and trustworthy remains the bedrock upon which AI, automation and digital transformation depend. The tools may change and the expectations may rise, but that fundamental requirement stays the same.
And if all of that doesn’t excite you, we have some new AI features which will revolutionise your business…
References
