Dominic Wellington of SnapLogic warns of an “orchestration” wall that could lead to AI becoming yet another expensive, ungoverned silo, costing leaders millions in the year ahead
Everyone expected 2025 to be the moment of truth for AI, when projects would start being deployed in production and deliver real-world impacts. Instead, we saw a spate of stories about huge investments in AI and little to show for it, a failure sparking justifiable fears of an overinflated market and a looming bubble.
The reason for this shortfall is clear: in 2026, enterprises will hit an AI governance and orchestration wall. Early AI platforms are being scaled without proper governance and cross-enterprise orchestration. Most importantly, it’s not a lack of AI intelligence causing this failure, but the inability to operate effectively within a regular enterprise environment. The best agentic AI platform is no good if it can’t get seamless, governed access to the legacy apps, mission-critical systems, and regulated data that business functions run on today.
Crucially, as these siloed agents collide or simply fail to interoperate, companies are being forced into costly iterative rewrites just to make systems functional, secure, and compliant. The blunt reality is that the hardest mile in AI isn’t getting models to run – it’s cleaning up the mess they leave behind when they are left to their own devices. Without a foundational layer of cross-enterprise orchestration, AI will continue to fall at the last, most critical hurdle of execution and longevity.
The challenge of the bottleneck
To understand where this challenge has come from, we need to consider which factors may have led to silo-ification in the first place, and how they can be overcome. Much as we saw with the previous revolution around cloud computing, true benefits are not realised from simply doing the same old things slightly faster or more cheaply. Achieving transformational benefits requires, well, transformation: embracing the new possibilities of AI.
Equally important, though, is a clear-eyed recognition that AI is not magic, and cannot deliver that transformation on its own. Companies that tried to treat cloud computing as a new way to procure servers, which they then treated as if they were physical servers or on-premises VMs, did not get the benefits of cloud. Those benefits went to the companies that embraced new modes – dynamic and responsive infrastructure and services at all levels of the stack, and the operational and even financial models to match. We will see the same mechanisms play out with AI, compounded by the rising tide of unmanaged AI agents. Without centralised governance, those early, siloed experiments become technical debt overnight, guaranteeing the expensive rewrites and confusion we are now seeing.
Setting up for success
My biggest recommendation for businesses to avoid these conundrums is to not focus on the technology. If your project is a good fit for AI, that will become obvious once you have properly mapped out how it operates today, how it should operate, and how to know when success has been achieved. Without an answer to those questions, no mandate from above to “use AI” can succeed. By all means, do use AI to help you work out what you should be doing. Some of the new research modes can help identify and explore possibilities and build out the business case for implementing them. That way, AI can provide benefits before you even break ground on the implementation part of the project – not least by helping you avoid dead-ends or technological white elephants that are delivered for their own sake.
Cross-enterprise orchestration
Much as human users rarely spend their entire day working in one single software tool, it is simply impossible to achieve high-value business goals with AI deployments that operate entirely within a single silo. Instead, achieving those all-important business goals will require both users and AI agents to orchestrate across many different systems and data sources.
Fortunately, there’s already a deep well of wisdom about such orchestration to draw from, built from years of best practice. The reason these known formulas have not been implemented as pervasively as we might wish is that all too often proper orchestration was seen as a “nice-to-have”. The enormous compliance requirements of AI, and the outsized returns it promises, now make it an absolute must-have to break down the barriers that prevent such orchestration.
New, agent-specific integration models such as MCP and A2A are part of the conversation, but are not sufficient on their own. Concerns over governance, risk, and compliance cannot be addressed with purely technological approaches – they require organisational action. The solution is to bring stakeholders into the conversation early, so they are defining the required architecture, not just flagging problems after deployment.
To succeed in 2026, firms must mandate that AI is deeply integrated into the enterprise architecture and the strategic conversations around it, instead of allowing it to become yet another expensive, ungoverned silo. This will be the only way for leaders to safeguard their investment and deliver on AI’s moment of truth. Get it right, and businesses can achieve enterprise-wide scalability and efficiency that reaps the rewards of AI through measurable ROI and a significant competitive advantage.
About the author
Dominic Wellington is the Director of Product Marketing
for AI and Data at SnapLogic.
Further reading
This article was first published in Business 5.0.