Artificial intelligence systems capable of making autonomous decisions are set to play a growing role in water utility operations, although human oversight will remain essential for critical decisions, according to a report by water technology company Xylem Vue.
The application of artificial intelligence to water processes and services is evolving rapidly as Agentic AI architectures are beginning to establish themselves as one of the main drivers of transformation in water operations, stated the expert in its Water Technology Trends 2026 report.
These autonomous agents perceive their environment, make decisions, and act to achieve objectives. They are enabled by Model Context Protocols (MCPs), which allow AI models (such as ChatGPT) to connect securely and seamlessly to external tools, data, and systems, it stated.
Large language models (LLMs), artificial intelligence models trained on vast amounts of text to understand and generate human language, can now be deployed as coordinating agents that interact with business systems in a structured and governed way, rather than relying on isolated analyses, rigid dashboards, and monolithic models trained to perform individual tasks.
LLM-based agents use MCPs to gain controlled access to operational data, analytical services, and execution capabilities, it added.
David Torres, the AI Product Manager at Xylem Vue, said MCPs define standardised mechanisms that enable agents to identify available tools, retrieve contextual information, invoke analytical processes, and, at times, trigger operational actions.
“This ensures that reasoning and execution occur within auditable, secure, and domain-specific parameters, which are essential in critical infrastructure such as the water sector,” he stated.
Agentic AI architectures
Xylem Vue said the so-called "agentic AI" architectures, powered by large language models (LLMs) and Model Context Protocols (MCPs), are emerging as a key technology for managing water infrastructure by enabling AI systems to securely access operational data, analytical tools and business systems.
The technology allows water utility operators to interact with systems using natural language rather than relying on fixed dashboards or reports, enabling AI to retrieve, analyse and present operational data in real time to support decision-making, it added.
AI approaches enabled by MCP introduce a fundamentally different paradigm from traditional software systems in that they allow for operator-driven customisation. Instead of relying on predefined dashboards, reports, and KPIs, operators can express their analytical needs and objectives in natural language.
These “AI agents” convert these requests into structured processes for retrieving, analysing and visualising data in real time, said the water technology company in the report.
In addition, they can retrieve, combine and format data on demand, presenting them in the most appropriate format for decision-making, such as time-series charts showing KPI trends and predictions; summary tables and lists of anomalies classified by different criteria; and customised thematic maps that apply geoprocessing techniques and comparative analyses between DMAs and assets across different time windows and operating conditions.
Utilities can define recurring analytical workflows, such as daily or weekly reports highlighting key events, anomalies, and KPIs, in addition to ad hoc queries.
This flexibility reduces the friction between data and decision-making, enabling teams to focus on interpreting results and prioritizing actions, rather than gathering information and managing multiple systems, it added.
Human oversight essential
However, despite advances in AI agents, risks related to security, reliability, and liability persist, especially in critical infrastructure such as water systems, where a single mistake can have serious consequences. For this reason, “not all decisions should be automated and fully delegated to these systems,” stated Torres.
MCP-based agent architectures incorporate human-in-the-loop governance, enabling the definition of which critical decisions require human validation right from the design phase. This is why purpose-built MCP agents for the water sector are so important.
They can be designed to analyse conditions and generate recommended actions; clearly explain the reasoning and evidence behind each recommendation; and request explicit human approval before running certain predefined operations,, he added.
According to the report, common use cases include approving interventions, operational changes, emergency activations, and communications, while ensuring that expert judgment remains at the center of decision-making, supported by advanced analytics.
This human-AI integration facilitates gradual adoption, starting with advisory systems and increasing automation as trust and organizational maturity grow. Furthermore, human decisions become key data for evaluating and improving AI performance.
Regulatory constraints are translated into security controls within servers and MCP tools. LLMs provide some of the intelligence, but execution depends on the specific design of the MCPs for the water sector.-TradeArabia News Service