The Integration and AI Landscape

In this newsletter we examine the importance of Integration, and how AI fundamentally changes the Integration landscape.

Integration is Critical

The days of large monolithic systems that live in a bubble are over. Software these days is composed of lots of parts which need to talk to each other. To be truly valuable, they need to also exchange information with other systems. Information is the lifeblood of any organisation. It is a critical input in decision making for both you and your customers. Without it processes don't function.

As technology has evolved, the need for integration, to deliver information between systems, has continued to grow exponentially. Over the years computing has transformed from mainframes to personal computers to the internet, to cloud computing. At each stage new mechanisms and approaches to exchanging data have been created. We've progressed from paper, through magnetic media (tapes & disks), modems, websites, a range of obscure communications protocols, web services, APIs, event streaming and all sorts of variants and cross overs in between.

AI has thrown all sorts of new considerations into the mix. AI is no longer just a chatbot that you talk to. It is a powerful tool able to hook into other systems to undertake tasks, take on board information, act on your behalf, recommend, and even make decisions.

A Brief History

I'll make a small detour here. It is worth noting that the notion of Artificial Intelligence (AI) has been around for a long time. It was certainly something that was taught when I did Computer Science in the 1990's, and wasn't new even then. At a very high level it can be thought of as a collection of approaches to be used when asking machines to solve problems. What tends to be referred to as AI now, is based around the concept of neural networks. These are pieces of software with logical pathways that are reinforced (more likely to be used) based on exposure to patterns, a little like the way the brain functions. Decisions therefore become driven by statistical probability (depending on the weight on those pathways), but with vast amounts of permutations possible.

It is only recently that large language models (LLMs) have made AI accessible to the masses, in a way that results in it being talked about in daily conversation and mainstream media. AI research had largely been the domain of researchers in Universities and organisations solving specific problems. Then Open AI came along with ChatGPT, quickly followed by the likes of Anthropic (with Claude), and Google (with Gemini). These large players provide what are known as frontier models. They are the extremely large conversational AI models trained across vast amounts of data, that have a broad knowledge base and require massive amounts of hardware to run. There are others too - but hopefully you get the idea. We'll also park conversations about the substantial energy and water use for another day.

There used to be a time when family members would say to me that the computer wasn't doing something for them that was working for someone else. I'd point out that it was all 0's and 1's, so it was probably something rational, resulting in the behaviour they were seeing. In computer science this is referred to as determinism - having the same specific outcome based on a particular set of inputs. It is the way computer scientists, and software developers are historically trained to think. Modern LLMs (we'll just call them AI from now on) turn things on their head. Despite running on silicon chips using 0's and 1's, they are for all intents and purposes, non-deterministic. That means that given the same set of inputs, you can't guarantee you are going to get the same answer, which can sometimes be quite frustrating.

I've heard it said that people managers are often better placed to work with AI than those with a background in managing people. I'm not sure it's anywhere near that simple, but I think the key point is that providing adequate instruction, context, and checking in on progress are all relevant, just as they are when working with people.

What about AI Agents?

Cue the notion of AI Agents. A lot of people have very different ideas about what an Agent is. For the purposes of this conversation, we'll consider them to be a piece of software (or code if you prefer) that executes a loop to do a job, working out each next step using AI (on each pass through the loop), and performing actions along the way, until that job is completed.

The real power is in having them perform tasks, which means they need to talk to other systems i.e. integration is required. There are a number of challenges with that. Agents are being spun up to do jobs all over the place. Some are well thought out and structured - others less so.

Agents and Integration

Agents are often connected to other systems using whatever means are available - for example existing APIs, filesystems, and browser extensions. Many of these (yes even APIs) have often been created with a different purpose or workflow in mind.

Cue a new standard - Model Context Protocol (MCP). Its intended purpose is to solve this by providing a mechanism for AI agents to interact with systems to perform tasks or obtain information that are beyond its training or capabilities. It is provided by an MCP Server - which is software that acts as an integration layer (like a bridge) between the agent and the target system. It utilises an appropriate set of tools to interact with the target system. Those tools may include the mechanisms mentioned earlier.

In essence, it gives agents a way to do things they can't do for themselves, either because they don't know how, or they simply don't have the capability to perform those actions. Importantly it affords agents the flexibility they need to solve problems. MCP was originally a stateful protocol, but moved to a stateless model with the July 2026 spec update — so this is still bedding in across the ecosystem. It's intended to work across multiple agents and AI platforms. The great thing about MCP is that it has genuinely wide adoption — OpenAI and Google DeepMind have both built support for it, and in December 2025 Anthropic handed MCP over to the Linux Foundation's Agentic AI Foundation, backed by AWS, Google, Microsoft, OpenAI, Bloomberg and Cloudflare. So there's no VHS versus Betamax argument going on. The end result is that AI agents can be more capable and efficient in the way they do things.

The Integration Impact

Over the past 20 years or so we have moved from a world where we had fewer larger systems, talking to each other over (hopefully) well-defined interfaces, through to splitting those systems into reusable parts (microservices for the techies out there), throwing in cloud based software services, and now we've thrown agents into the mix. The result is an exponential growth in the amount of integration going on.

Many larger organisations that are mature have the notion of Integration Governance that is important for ensuring a structured approach to concerns such as reusability, resilience, privacy and security. Those needs didn't suddenly vanish overnight. The move to AI begs many questions, many of them directly related to integration concerns: How do you feed information to the AI, for effective decision making? Where is my data going? Do other people see my data? How do I pull together information from my different systems? How can I get AI to do more so I can free up time to focus on my core business? What's the on-going role of my current integration platform? And the list goes on…

The Role of AI and Integration for Smaller Organisations

AI has fundamentally shifted the playing field, in what is achievable, but also in terms of the way systems talk to each other, and source reliable information. I believe that the effective Integration of AI will be critical for competitive advantage and long term viability of all organisations. There is a lot of talk regarding the potential for AI to improve efficiency and lift productivity. Those benefits don't just belong to large organisations with larger head counts and deeper pockets.

I'm keen to help small to medium sized organisations get started on that journey. My strong opinion is that it is a critical tool in the tool box, with something to offer every organisation. But like power tools on a building site it needs to be treated with respect, used correctly, and applied in the right situations. Not every situation requires a nail gun. Just because you can, doesn't mean you should.

Integrating information from different systems to facilitate better customer outcomes and quicker decision making is a no-brainer. But there are guard rails that need to go with that. AI is well suited to situations where there is reasoning (thinking) involved. It is not best suited for well-defined repeatable processes where consistency of outcome is important. However, you might use it to figure out how to define that process, or even to build a tool that implements it. Always remember that AI and humans work best together. Ultimately you still need to be responsible for how and where you apply AI, what it creates and how you apply that.

Wrap Up

Integration and AI are more than just tech trends—they're the new foundation for how we work. Whether you're a small business or a growing organisation, the ability to connect your systems intelligently is a massive competitive advantage. But as with any powerful tool, success comes from applying it thoughtfully, maintaining clear guardrails, and remembering that AI is at its best when it's partnering with, not replacing, the human touch. Thanks for reading — I'd love to hear your thoughts, so reply any time at feedback@queleon.io, and if you haven't already, please subscribe here to receive future issues directly in your inbox.

© Queleon Limited 2026. All rights reserved.