Agentic AI Takes The Next Step In Addressing The Water Sector's Challenges

Operational resilience, regulatory auditability and knowledge sustainability are the three areas where agent architectures are beginning to deliver differentiated value, according to Xylem Vue.

Governance is no longer an add-on; it is embedded in the layer that controls the model’s access to a utility’s operational systems.

The water sector faces a complex reality: ageing infrastructure, insufficient investment to renew and expand networks, and a generation of professionals approaching retirement. The American Water Works Association estimates that more than 30% of water-sector professionals in the United States will retire by 2030. These challenges are compounded by climate volatility and growing pressure to deliver services that are more resilient, transparent and efficient.

Against this backdrop, generative AI is taking on a leading role by unlocking value in areas whose cross-cutting, unstructured and context-dependent nature had previously placed them beyond the reach of traditional analytics and machine-learning tools.

Agent architectures add a reasoning layer that orchestrates these models, makes them accessible through natural language and extends their reach to processes that previously remained outside the scope of digitalisation. “This is the frontier that is shifting in 2026,” says Jaime Barba, Head of Xylem Vue. “We are not replacing anything that already works; we are opening a second front on challenges that, until now, depended exclusively on human judgement.”

Operational Resilience In The Face Of Multidimensional Crises

According to the report “Water Technology Trends: Revolutionising Water Management”, published by the Xylem Vue platform, the challenges ahead are increasingly focused on business resilience. The crises facing utilities, whether climate-related, cyber-related or infrastructure-related, rarely emerge within a single system. A severe event can simultaneously span SCADA, GIS, CMMS, alarm systems and emergency protocols. Until now, operators have had to mentally piece this information under intense time pressure.

Agent architectures based on MCP enable an operator to ask a question in natural language and receive an integrated response across multiple domains, with the reasoning traced step by step. Contingency plans cease to be static documents and become dynamic contexts that can be consulted in real time, while the operator always retains control over any resulting actions.

Another key challenge is regulatory performance and auditability by design. Regulatory reporting obligations are multiplying: in Europe, through the implementation of the EU AI Act; in the United States, through EPA and CISA cybersecurity frameworks and several state-level inquiries into the use of AI; and across the sector, through specific water-quality and cybersecurity regulations. Automating regulatory monitoring, report generation and multilingual alerts reduces the administrative burden, but the real differentiating value lies elsewhere.

“Every step in the agent’s reasoning is recorded: which tools it consulted, which data it used, which recommendation it issued and what a human approved,” explains David Torres, AI Lead at Xylem Vue. “This traceability, embedded in the architecture by design rather than added retrospectively, is what turns an AI system into a system that can be audited by a regulator, an internal audit team or a post-incident investigation.”

Knowledge As A Strategic Asset

Two seemingly different challenges, the large-scale retirement of professionals with decades of tacit knowledge and the growing need to explain technical decisions to regulators and citizens, share a common technical solution. An AI agent connected to a utility’s knowledge base and to the sector’s accumulated expertise can propose the action that a senior operator would recommend, explain its reasoning and adapt that explanation to the audience. This knowledge, including procedures, operational history and decision-making criteria, always remains under the utility’s ownership and control. It is neither transferred to the technology provider nor used to train third-party models.

For a junior operator, this shortens the learning curve and mitigates the loss of institutional knowledge. For a citizen or regulator, it turns an opaque decision into one that can be explained and justified. In both cases, the underlying technical capability is the same: reasoning based on water-sector-specific context, with complete traceability of the rationale behind each conclusion.

Governance: The Pillar Already Embedded In The Architecture

Data governance is another challenge that must be addressed. The adoption of generative AI by utilities requires a rigorous approach to governance. In the model advocated by Xylem Vue, governance is no longer an aspirational objective or an add-on introduced after deployment. Instead, it is embedded in the layer that acts as an intermediary between three elements: the operator submitting the query, the language model providing the reasoning, and the utility’s operational systems, such as SCADA, asset management and work-order platforms, where data resides and actions are executed. This intermediary layer, which Xylem Vue calls a “water-specific harness”, orchestrates which tools the model invokes and in what order, returns the response to the operator in the appropriate format, and determines what the model can access, what it can do, under which conditions and with what level of human oversight, applying criteria specific to the water sector.

In practice, this translates into permissions by asset, area and role; a distinction between read-only queries and actions that modify the system; mandatory human validation for consequential operations; a complete audit trail of every interaction; and the ability to reassess the system whenever a prompt, tool or data source changes. All of this is aligned with the requirements of the EU AI Act.

“What matters is not which model the utility chooses, but the water-specific governance layer that determines whether that model is safe to operate on critical infrastructure,” Torres says. “Xylem Vue provides that layer as the technology provider, but the data, operational knowledge, contexts and configuration decisions always remain in the utility’s hands, regardless of the level of AI adoption the organisation chooses.”

Xylem Vue is a secure, integrated, and agnostic analytics and software platform that can capture data from any source, including legacy solutions. The platform enables water utilities to maximize investments already made in existing technologies while moving further along their digital journey and breaking down data silos to provide a holistic, 360-degree view of their system. Xylem Vue offers a broad portfolio of modular applications, helping utilities unlock the true power of their data and solve their most pressing water cycle challenges more efficiently.