Why is infrastructure stagnating?

According to the EY report The intelligence layer: how agentic AI can connect the infrastructure industry, the main obstacle for infrastructure is not a lack of technology, but fragmentation of systems, data, and stakeholders. Current project management models require manual coordination and periodic reporting, which creates delays and blind spots in decision-making.

Data is locked in separate silos — from construction companies through material suppliers to regulatory authorities. This state limits visibility of the entire project and prevents effective collaboration. EY estimates that losses caused solely by the need for rework reach hundreds of billions of dollars annually. Investment in digital tools has grown in recent years, but their productive benefit remained local — tools improved individual tasks, but not the overall connectivity of the project.

What is agentic AI and how does it work in practice?

Agentic AI represents an evolution of generative artificial intelligence. While classic AI models respond to queries and generate content, agentic AI acts independently — analyzes data in real time, plans steps, coordinates tasks across systems, and can learn from the results of its decisions.

In the context of infrastructure, it functions as an intelligent layer above existing systems. It does not try to replace ERP platforms, BIM models, or project managers, but connects them. Agents continuously collect data from various sources, uncover patterns, and provide project teams with concrete, immediately usable information. Humans remain responsible for strategic decisions and expert judgment, while AI handles routine administration and coordination.

„Romanian infrastructure ambitions require a fundamental change in how we transform investments into results. Agentic AI offers a practical way to connect fragmented data and decisions across projects, helping organizations move faster, reduce rework, and build trust through strong governance. The real benefit lies in clearer insights, not just automation," said Aurelia Costache, partner for consulting and AI lead at EY Romania, in an official statement regarding the report.

Where will it specifically help?

Reducing rework costs

Up to 15 percent of the budget for large infrastructure projects disappears into fixing errors that arose due to poor communication between teams or outdated data. Agentic AI can significantly reduce this share by uncovering discrepancies before they are reflected in physical construction. The system continuously checks the consistency of project documentation with real-time data from the field, supplier invoices, and regulatory requirements.

Faster decision-making and reporting

Current project teams often spend dozens of hours per week collecting data from various spreadsheets and systems for regular reports. Agentic AI automates this process and provides continuous overview of project status. Managers receive risk alerts immediately, not only at monthly evaluations.

Resource optimization

Thanks to predictive capabilities, agentic AI can better plan the use of workforce, materials, and machinery. This is crucial at a time when qualified workers in construction and logistics are lacking worldwide. The same team can handle a larger volume of work through better coordination without the need to increase personnel.

Five pillars of successful implementation

EY emphasizes in its report that deploying agentic AI in infrastructure is not merely a technological matter, but requires a robust governance framework. Without it, there is a risk of losing control over autonomous systems and undermining the trust of stakeholders. The report identifies five critical areas:

Accountability and legal liability — it is necessary to clearly define who bears responsibility for decisions made by an agent. In case of an error, it must be clear whether the algorithm creator, system operator, or project manager is responsible.

Transparency and explainability — agents must be able to justify their recommendations. A black box that issues orders without explanation is unacceptable in critical infrastructure.

Data sovereignty and business confidentiality — infrastructure projects include sensitive information about costs, contracts, and security measures. Agentic AI must process this data in compliance with regulations and prevent unauthorized access.

Human oversight and competence — autonomy must not mean absence of control. People must have the ability to intervene at any time, override an agent's decision, and remain the final arbiters.

System resilience — critical infrastructure must be able to function even during AI system outages. Backup mechanisms and manual modes are essential for ensuring continuity.

What is the impact for the Czech Republic and Europe?

While the EY report is based on global data, its conclusions are fully relevant for Czech and European conditions as well. The Czech Republic faces long-term problems with the completion of transport infrastructure, where delays and budget overruns are not the exception. A similar situation prevails throughout Europe — projects such as high-speed railways or energy transformations suffer precisely from the fragmentation of management and slow decision-making processes.

Agentic AI is not a proprietary product of a single company, but a concept applicable to various platforms. In practice, organizations can integrate it through cloud solutions from Microsoft (Azure AI Agent Service), Google (Vertex AI Agent Builder), or Amazon (AWS Bedrock Agents). These services are also available to Czech companies and institutions, with the price depending on the volume of computing resources used and the number of integrated systems. Typically, this is pay-as-you-go subscription or enterprise licenses, whose monthly costs can start at hundreds of euros and climb to tens of thousands for extensive infrastructure projects.

The European context additionally emphasizes AI regulation — the EU AI Act demands a high degree of transparency and human oversight for systems used in critical infrastructure. This may on one hand be an obstacle to rapid deployment, but on the other hand creates a solid framework for the safe introduction of agentic AI precisely in areas such as transport, energy, and water management.

For ordinary companies and state institutions in the Czech Republic, this means they do not have to wait for a miraculous tool — the key is to start by connecting existing data sources and gradually introduce autonomous agents for specific, well-defined tasks. The path to full utilization of agentic AI leads through thorough data preparation, clearly defined processes, and strong governance — not through replacing human decisions with machines.

Frequently asked questions