Integration in an AI World

Having attended Integrate for many years, I have always regarded it as one of the standout events in the integration community. It consistently provides practical insight into where our industry is heading, and this year's conference was no exception.

What immediately stood out was the prominence of Artificial Intelligence across the agenda. Whether sessions focused entirely on AI or explored how it is reshaping established integration practices, one message was clear: AI is no longer an emerging capability; it is becoming an integral part of modern integration architecture.

Before the conference began, one session had already made its way onto my "must watch" list: Integration in an AI World, delivered by Dan Toomey.

Having followed Dan's presentations over the years, I expected a session rich in practical guidance, and it certainly delivered. Dan has a rare ability to explain complex architectural concepts in a way that is both relatable and immediately applicable. Rather than focusing on AI hype, he explored what the rise of autonomous AI means for enterprise integration and, more importantly, what organisations should be doing today to prepare.

Dan structured the session into three acts, each building on the last.

Act One – The Shift in API Consumers

The opening act challenged one of the assumptions many of us have held throughout our integration careers.

For more than two decades, APIs have been designed primarily for software developers. Documentation, developer portals, and intuitive REST interfaces have all been optimised with human consumers in mind.

Today, that landscape is changing.

One of the strongest messages from the session was that enterprise APIs now have two distinct consumer groups: human developers and AI agents.

While developers continue to design and build applications, AI agents are increasingly discovering, selecting, and orchestrating APIs autonomously to complete business tasks. Unlike developers, they do not browse documentation or developer portals. They consume OpenAPI specifications, JSON schemas, metadata, and structured contracts to determine which APIs best satisfy a particular objective.

This shift has significant implications for how we design APIs.

An API that is poorly described, inconsistently modelled or unreliable may not fail, it may simply never be selected by an AI agent.

For me, this was one of the most important takeaways from the session. Being "API-first" is no longer enough. Organisations should also be thinking about becoming agent-ready, ensuring their APIs are:

• Machine discoverable

• Semantically rich

• Predictable

• Reliably governed

• Observable

These characteristics are rapidly becoming as important as the API functionality itself.

Act Two – Integration Becomes the Execution Layer

The second act was, in my opinion, the highlight of the session. Dan described integration as the execution layer for AI.

Large Language Models can interpret requests, generate plans and reason through complex problems, but they only create business value when they interact with enterprise systems. APIs and integrations are what transform AI-generated intent into significant business outcomes.

That perspective fundamentally changes the role of integration. Rather than simply connecting systems, integration platforms serve as the trusted execution environment through which autonomous AI interacts with enterprise applications.

Dan illustrated this with a simple but powerful example.

An undocumented API change caused an inventory field to disappear. The AI agent interpreted the missing value as zero stock and automatically placed an urgent order worth hundreds of thousands of dollars.

Importantly, this was not an AI hallucination. It was not a poor prompt. It was an integration failure.

The example perfectly demonstrated that AI does not remove the need for good integration practices. Instead, it amplifies their importance.

As AI operates at machine speed and machine scale, weaknesses in integration architecture become magnified just as quickly.

That places even greater emphasis on:

• API governance

• Contract management

• Versioning

• Observability

• Security

• Reliable integration patterns

Good integration increases the value AI can deliver. Poor integration increases operational risk.

Another point that resonated with me was Dan's observation that organisations with mature API-first strategies already possess a significant advantage. Businesses that rely heavily on manual processes or legacy applications without accessible APIs face what he described as a "wrapping tax", the additional effort required to expose existing capabilities before AI can effectively consume them.

Act Three – Designing for an Agent-Ready Future

The final act moved away from concepts and into practical guidance.

Rather than introducing entirely new design principles, Dan demonstrated how many existing integration best practices become significantly more important when AI becomes a consumer of enterprise APIs.

Several recommendations stood out:

• Describe APIs with intent
  Tool names, endpoint descriptions, and metadata should clearly communicate purpose, expected behaviour and constraints.

• Publish rich, machine-readable specifications
  OpenAPI definitions, JSON Schemas and consistent contracts become essential for autonomous discovery.

• Design for composition
  Smaller, focused APIs enable AI agents to orchestrate business capabilities far more effectively than large monolithic endpoints.

• Invest in observability
  Comprehensive tracing, telemetry, and correlation become critical when debugging autonomous behaviour rather than traditional application code.

• Build governance into every integration
  Security, least-privilege access, rate limiting and policy enforcement should be considered foundational requirements for agent-enabled architectures.

I also found Dan's discussion around Azure API Center particularly interesting. While API Management often receives most of the attention, API Center has the potential to become an increasingly important capability for cataloguing, governing, and making APIs discoverable for both developers and AI agents.

The Valorem Reply Perspective

One of the strongest messages I took away from the session was that enterprise APIs are entering a new era. They are no longer designed solely for developers.

Increasingly, they are being consumed by autonomous systems that discover, select, and orchestrate business capabilities dynamically.

That does not diminish the role of human developers. Instead, it introduces a second consumer that has fundamentally different expectations of our integration landscape.

For organisations adopting AI, this creates both an opportunity and a responsibility. The opportunity lies in enabling AI to automate increasingly complex business processes.

The responsibility lies in ensuring the underlying integrations are secure, observable, governed and resilient enough to support that autonomy.

At Valorem Reply, this aligns closely with the way we help our clients modernise their integration platforms.

Becoming agent-ready is not about replacing existing integration strategies. It is about extending proven architectural principles to support a future where humans and AI work together.

That means designing APIs that are easily discoverable, consistently governed, and capable of supporting autonomous execution without compromising enterprise control.

As AI continues to evolve, organisations with strong integration foundations will be best positioned to adopt new capabilities with confidence.

For me, Dan Toomey's session was one of the highlights of Integrate 2026 because it reinforced a simple but important truth:

Integration is no longer simply the plumbing behind digital transformation. It has become the trusted execution layer that enables enterprise AI to deliver significant business value.