For years, we have been piling up tools that are supposed to streamline work: messaging systems, CRM, ERP, collaborative platforms, etc. Each new need corresponds to a new tool that is added to the others, often independently.
Yet despite this, the daily lives of employees remain fragmented between tools, workflows experience load breaks, data is scattered, and coordination is sometimes complex and always complicated.
In reality, it would be more accurate to say that these tools improve work to a certain extent but rarely streamline it, and may even have the opposite effect.
I have already addressed this problem on several occasions: the experience of tools is designed vertically within each tool, whereas the employee’s journey through their work is cross-functional and moves from tool to tool (What (digital) workplace experience for your employees ?and A complicated IT experience. Irritant #7 of the Employee Experience).
Some see the emergence of generative AI as a magic wand that will solve all problems, including this one. AI would be able to link applications and data, eliminate fragmentation, and allow users to do all their work from a single, conversational interface, without having to worry about the tools behind it.
This is a popular idea that some people, whom I would describe as dreamers or idealists, are taking to extremes by predicting that AI will make applications disappear. I tend to agree with those who think that applications will become “headless” but will not disappear altogether ([FR]Personal AI agents, headless SaaS… what the future holds).
But this reminds me of a subject that used to drive me crazy back in the distant past when I was arguing for interoperability between business tools and social collaboration platforms, and I remain convinced that this lack of interoperability contributed greatly to the failure of the latter (The rise and fall of enterprise social networks). I think we are back to the same problem and that one question remains largely underestimated: is this really a question of AI, or are we simply continuing to sidestep the real issue, which is interoperability?
In short :
- The accumulation of digital tools creates fragmentation of tasks and data.
- Interoperability is primarily an organizational challenge, beyond the technical.
- AI depends on access to data and integration capabilities, without resolving fragmentation.
- Overlays such as LumApps improve the experience but remain dependent on the limitations of business systems.
- APIs and MCPs are essential for activating data within a framework of good governance.
Fragmentation is structural
The digital workplace was never designed as a coherent system but was built up gradually, with each business unit adding its own tools according to its priorities. A world of opportunistic development has produced an environment where dozens of parallel systems coexist, rarely integrated or synchronized with each other.
A good example of this is the exponential growth of certain tools such as Lumapps and Powell, which have been added to this application puzzle to offer a more coherent user experience, but this is a subject I will return to later.
But this fragmentation is not only technical. It is often the result of a siloed, fragmented and therefore complicated organization, which contributes to reinforcing this phenomenon and its effects (The organizational complication: the #1 irritant of the employee experience): friction at every level of the business, progressive burnout of teams, who spend more time compensating for the system’s flaws than doing their jobs and actually producing something. This is true in real life, and it is therefore also true in tools.
Interoperability: a key issue that is poorly addressed
This situation is not the result of a lack of tools, but rather an inability to make them talk to each other. Interoperability remains the weak point of most digital workplace strategies.
However, it comes into play at several levels:
- Technical: without open and maintained APIs, systems remain siloed.
- Functional: without consistency or even “synchronization” of business processes, the organization is not aligned.
- Semantic: without shared repositories, data circulates without a shared meaning.
Yet we continue to approach interoperability as an IT integration problem when in reality it is primarily a matter of organizational design or even work design. Workflow design, business structuring, and governance often determine data design, and as long as these issues remain unaddressed, the stacking will continue.
IT and organization follow and reinforce each other: if one thinks in silos instead of synergies, the other will follow.
AI and fragmentation: a miracle solution?
The arrival of generative AI is currently fueling a kind of fantasy. Since systems are not integrated, it would seem that all we need to do is entrust AI with the task of finding and aggregating scattered information, or even acting on it by controlling the tools that contain it. And in some cases, this approach works: AI can find, summarize, and produce content from scattered sources.
But this ability remains conditional on three factors: access to data, the quality of sources, and the ability to control the tools. AI does not create interoperability but depends on access to information and tool functionality. When systems are closed, there are no APIs or only limited APIs, and access rights are unclear, AI cannot work miracles.
Above all, another form of fragmentation must be taken into account: every publisher now integrates its own AI co-pilot into its tools (Generative AI in the enterprise: a silo breaker or just another layer in an already complex IT landscape?) and does not want other AI systems to be able to draw on its data, except in rare cases where non-competing publishers decide to share AI or something close to it (Workday Salesforce Partnership: Teaming Up For Enterprise AI). Dispersion is not solved, it is simply moved elsewhere, and after application fragmentation comes assistant fragmentation, with each assistant confined to its own perimeter, unable to cooperate with others. This multiplies cognitive silos without changing the structure of the problem, or even allowing things to get worse.
Of course, there are businesses that break free of these limitations with in-house developments, but this comes at a price that few can afford.
Software overlay on the digital workplace: a great idea, but not a magic wand
Earlier, I mentioned the success of solutions that provide an overlay for the multiple applications that make up the digital workplace and, beyond improving user experience consistency, are beginning to deliver headless interaction with these tools.
The idea is brilliant, and I’ve seen demos that have impressed me, but these publishers are not magicians.
For example, if I understand the logic behind Lumapps (a superb solution, by the way), its AI assistants improve access to internal information (smart search, content suggestions, editorial assistance) and facilitate navigation to certain business actions via API integrations, but they do not directly execute business transactions (e.g., requesting time off, approving an absence). These operations are still handled by the transactional engines of business applications(Workday, Salesforce, etc.) or by specific integrations developed as an overlay. LumApps’ AI acts on the experience and content layer, not on the transactional business logic, but this is already a big step forward.
However, they cannot go beyond what APIs allow, and certain needs will require specific developments.
API and MCP: two complementary layers
To take this further, if we want AI to play the role of experience enabler that we would like to see it play, two things are necessary.
APIs open up access to data
APIs remain the foundation. They make data from business systems accessible, synchronize statuses, and retrieve information useful to processes. Without robust APIs, AI models have only partial visibility into what is happening in applications and, therefore, into the life of the business.
MCPs orchestrate the use of AI
Model Connectivity Platforms (MCPs) add a second layer. Their role is to manage the coordinated use of multiple AI models. In a mature digital environment, it is not a question of relying on a single model, but on a portfolio of specialized engines: generalist models, internal business models, etc.
MCPs enable requests to be routed, multi-source responses to be composed, and the governance of AI usage to be managed. Where APIs open up access to data, MCPs organize the exploitation of AI capabilities on that data. One without the other is not enough.
Behind the technology lies governance
Beyond the technical layers, there is a real issue of governance. Technical and cognitive interoperability require the organization to have clarified:
- the structure and ownership of its data,
- the consistency of its business repositories,
- the rules for managing information flows,
- access rights and decision-making processes.
Without this governance, adding new technologies only produces superficial effects and can even make things worse.
Bottom line
Fragmentation remains the central problem of the digital workplace today. Contrary to what some would have us believe, AI does not solve this problem, but rather reveals it. Its success will not depend on the proliferation of AI co-pilots, but on the ability to design a coherent information and process architecture and make it all AI-enabled.
Image credit: Image generated by artificial intelligence via ChatGPT (OpenAI)
Related articles:
Mobile is more than mobile versions of desktops applications
What about the employee journey in the work environment ?
What (digital) workplace experience for your employees ?
Generative AI in the enterprise: a silo breaker or just another layer in an already complex IT landscape?
The intranet is dead, long live the intranet? A look back at the Arctus 2025 study
Generative AI in the workplace: revolution or illusion?
